JP2012516780A - A model-based neuromechanical controller for robotic legs - Google Patents

Abstract

  A model-based neuromechanical controller for a robotic limb having at least one joint is disclosed. The controller receives feedback data regarding the state of the robot limb and receives a state information from the finite state machine configured to determine the state of the robot limb and receives muscle state geometry A muscle model processor configured to determine at least one desired joint torque command or stiffness command to be sent to the robot limb using shape and / or placement information, reflex structure information and a neuromuscular model; A joint command processor configured to command biomimetic torque and stiffness determined by the processor to a joint of the robot limb. The feedback data is preferably provided by at least one sensor mounted on each joint of the robot limb. In a preferred embodiment, the robot limb is a leg and the finite state machine is synchronized to the leg walking cycle.

Description

RELATED APPLICATION This application claims the benefit of US Provisional Patent Application No. 61 / 148,545, filed January 30, 2009, the entire disclosure of which is incorporated herein by reference. Shall.

  No. 12 / 157,727 filed Jun. 12, 2008, which was filed Jun. 12, 2007, but has expired, US Provisional Patent Application No. 60 / 934,223 No. 11 / 395,448, 11 / 495,140, and 11 / 642,993, and are continuation-in-part applications. The entire disclosure of these US patent applications is hereby incorporated by reference.

  No. 12 / 608,627, filed Oct. 29, 2009 (which is filed on Dec. 19, 2006 but abandoned U.S. Patent Application No. 11 / 642,993). A continuation-in-part application, which claims the benefit of US Provisional Patent Application No. 60 / 751,680, filed on December 19, 2005, but has expired) U.S. Patent Application Nos. 11 / 395,448, 11 / 495,140, 11 / 600,291, and 11 / 499,853 (this was patented as U.S. Pat. No. 7,313,463, August 4, 2005. (Which claims the benefit of the filing date of US Provisional Patent Application No. 60 / 705,651, which has been filed in the past) and is a part of US Patent Application No. 11 / 395,448 This is a continuous continuation application. The entire disclosure of these US patent applications is hereby incorporated by reference.

  This application is also a U.S. Patent Application No. 11/11 entitled "Artificial human limbs and joints using actuators, springs and Variable-Damper Elements" filed March 31, 2006 by Hugh M. Herr, Daniel Joseph Paluska, and Peter Dilworth. This is a continuation-in-part of 395,448. U.S. Patent Application No. 11 / 395,448 is filed on March 31, 2005, but has already expired, the benefit of the filing date of U.S. Provisional Patent Application No. 60 / 666,876, as well as on August 1, 2005. It claims the benefit of the filing date of US Provisional Patent Application No. 60 / 704,517, which has been filed but has already expired.

  This application is also filed by Hugh M. Herr, Samuel K. Au, Peter Dilworth and Daniel Joseph Paluska on July 29, 2006, “An Artificial Ankle-Foot System with Spring, Variable-Damping, and Series-Elastic Actuator Components. Is a continuation-in-part of US patent application Ser. No. 11 / 495,140. U.S. Patent Application No. 11 / 495,140 claims the benefit of the filing date of U.S. Provisional Patent Application No. 60 / 704,517, filed on August 1, 2005, but has expired. This is a continuation-in-part of Patent Application No. 11 / 395,448.

  This application is also a U.S. Patent Application No. 11 entitled “Exoskeletons for running and walking” filed November 15, 2006 by Hugh M. Herr, Conor Walsh, Daniel Joseph Paluska, Andrew Valiente, Kenneth Pasch, and William Grand. This is a continuation-in-part of / 600,291. U.S. Patent Application No. 11 / 600,291 claims the benefit of the filing date of U.S. Provisional Patent Application No. 60 / 736,929, filed on November 15, 2005, but has expired, and This is a continuation-in-part of patent applications 11 / 395,448, 11 / 499,853, and 11 / 495,140.

  This application claims the benefit of the filing date of each of the above patent applications, and incorporates the disclosure of each of the above applications by reference.

STATEMENT REGARDING FEDERALLY SPONSORED RESEARCH AND DEVELOPMENT This invention was made with support from the US government under grant number VA241-P-0026 and was honored by the US Veterans Support Agency. The US government has certain rights in this invention.

  The present invention relates to a method for controlling a robot leg based on a walking model of a neuromuscular (a nerve and a muscle), in particular, control of an artificial joint and a prosthetic limb for use in a prosthetic device, a correction device, an exoskeleton device, or a robot apparatus. .

  Animal and human leg movements are controlled by complex neural networks. It was proposed at the beginning of the 20th century that the central pattern generator (CPG) forms the base of this neural network (Brown, TG, 1914. “On the nature of the fundamental activity of the nervous centers; together with ananalysis of the conditioning of rhythmic activity in progression, and a theoryof the evolution of function in the nervous system.J Physiol 48 (1), 18-46. G., Deliagina, T., Grillner, S., 1999. “Neuronal control of locomotion from mollusc to man. Oxford University Press, New York”).

  At present, CPG is composed of a layer of neuronal pools in the spinal cord (Rybak, IA, Shevtsova, NA, Lafreniere-Roula, M., McCrea, DA, 2006. Modeling spinal structures involved in locomotor pattern generation: insights from deletions during fictivelocomotion. J Physiol 577 (Pt 2), 617-639), the spinal cord via other neuron pools that transmit muscle synergies (other neuron pools channeling muscle synergies) Clin Neurophysiol 114 (8), 1379- 1389; provides flexional activity sufficient for locomotion even in the absence of spinal reflexes (Dietz, V., 2003. Spinal cord pattern generators for locomotion. Minassian, K., Persy, I, Rattay, F., Pinter, MM, Kern, H., Dimitrijevic, MR, 2007. Human lumbar cord circuitries can be activated byextrinsic tonic input to generate locomotor-like activity.Hum Mov Sci 26 (2), 275-295, Grillner, S., Zangger, P., 1979.On the central generation o Exp Brain Res 34 (2), 241-261; Frigon, A., Rossignol, S., 2006. Experiments and models of sensorimotor interactions duringlocomotion. Biol Cybern 95 (6), 607-627) It is believed that. Nonetheless, spinal reflexes are part of this complex neural network (Rybak, IA, Stecina, K., Shevtsova, NA, McCrea, DA, 2006. Modeling spinal circuitry involved in locomotorpattern generation: insights from the JPhysiol 577 (Pt 2), 641-658), selection of gait pattern, timing of extensor and flexor activities, and regulation of CPG output.

  By using a combination of a central pattern generator and a reflex with a regulatory effect, a lamprey (Ekeberg, O., Grillner, S., 1999. Simulations of neuromuscular control in lamprey swimming. Philos Trans R SocLond B Biol Sci 354 (1385 , 895-902), Salamander (Ijspeert, A., Crespi, A., Ryczko, D., Cabelguen, J. -M., 2007. Fromswimming to walking with a salamander robot driven by a spinal cord model.Science 315 (5817), 1416-1420), Cat (Ivashko, DG, Prilutski, B. I, Markin, SN, Chapin, JK, Rybak, IA, 2003. Modeling the spinal cord neural circuitry controlling cat hindlimb movement during locomotion. Neurocomputing 52- 54, 621-629; Yakovenko, S., Gritsenko, V., Prochazka, A., 2004. Contribution of stretch reflexes tolocomotor control: a modeling study. Biol Cybern 90 (2), 146-155; Maufroy, C, Kimura , H., Takase, K., 2008. Towards a general neural controller forquadrupedal locomotion. Neural Netw 21 (4), 667-681), and people (Ogihara, N., Yamazaki, N., 2001.Generation of human bipedal locomotion by a bio-mimetic neuro-musculo-skeletalmodel. Biol Cybern 84 (1), 1-11; Paul, C, Bellotti, M., Jezernik , S., Curt, A., 2005. Development of a human neuro-musculo-skeletal model for investigation of spinal cord injury. Biol Cybern 93 (3), 153-170) It has become an indispensable tool for studying different control methods. The importance of these models is determined by experiments (Pearson, K., Ekeberg, O., Buschges, A., 2006. Assessing sensory function in locomotor systems using neuro-mechanicalsimulations. Trends Neurosci 29 (11), 625-631). The goal is to reproduce the proposed CPG architecture and underlying reflex motion. However, little attention has been paid to understanding how such an architecture represents or encodes the principles of the gait mechanism.

  These principles suggest that little or no control is required for leg movement, as opposed to the complexity of neural networks already known. For example, gait (Alexander, R., 1976. Mechanics of bipedal locomotion. In: Perspectives in experimental biology (Ed. Davies, PS) Pergamon, Oxford; Mochon, S., McMahon, T., 1980. Ballistic walking. J. Biomech. 13 (1), 49-57) and running (Blickhan, R., 1989. The spring-mass model for running and hopping. J. of Biomech. 22, 1217-1227 McMahon, T., Cheng, G., 1990. The mechanism of running: how does stiffness couple with speed? J. ofBiomech. 23, 65-78). Studies have demonstrated that these mechanical systems are self-stabilizing when the mechanical systems are properly tuned (McGeer, T., 1990. Passive dynamicwalking. Int. J. Rob. Res 9 (2), 62-82; McGeer, T., 1992. Principles of walking and running.Vol. 11 of Advances in Comparative and EnvironmentalPhysiology.Springer-Verlag BerlinHeidelberg, Ch. 4; Seyfarth, A., Geyer, H. , Gunther, M., Blickhan, R., 2002. A movement criterion for running. J. ofBiomech. 35, 649-655; Ghigliazza, R., Altendorfer, R., Holmes, P., Koditschek, D., 2003 A simply stabilized running model. SIAM J. Applied. Dynamical Systems 2 (2), 187-218). Walking and traveling robots further demonstrated practical compatibility and control benefits derived from this principle (Raibert, M., 1986. Legged robots thatbalance. MIT press, Cambridge; McGeer, T., 1990. Passive dynamic walking. Int. J. Rob. Res. 9 (2), 62-82; Saranli, U., Buehler, M., Koditschek, D., 2001. Rhex: A simple and highlymobile hexapod robot. Int Jour. Rob. Res. 20 (7), 616-631; Collins, S., Ruina, A., Tedrake, R., Wisse, M., 2005. Efficient bipedal robots based onpassive-dynamic walkers. Science 307 ( 5712), 1082-1085). However, there remains an unresolved issue of how this and other principles of leg dynamics are incorporated into human motion control systems.

  The importance of this interaction between dynamics and motion control is recognized by neuroscientists and biomechanics researchers alike (Pearson, K., Ekeberg, O., Buschges, A., 2006. Assessing sensory function in locomotor systems using neuro-mechanical simulations. Trends Neurosci 29 (11), 625-631). For example, it is generally accepted that CPG forms the main drive circuit for motor action in locomotion (Grillner, S., Zangger, P., 1979. On the central generation of locomotionin the low spinal cat. Exp Brain Res 34 (2), 241-261; Dietz, V., 2003. Spinalcord pattern generators for locomotion. Clin Neurophysiol 114 (8), 1379- 1389; Frigon, A., Rossignol, S., 2006. Experiments and Models of sensorimotorinteractions during locomotion. Biol Cybern 95 (6), 607-627; Ijspeert, AJ, 2008. Central pattern generators for locomotion control in animals and robots: a review. Neural Netw 21 (4), 642-653) In 1969, Lundberg said that in the absence of that somewhat simple central input, the spinal reflex that relays information about locomotor dynamics forms the complex muscle activity seen in real locomotion. (Lundberg, A., 1969. Reflex control of stepping. In: The Nansen memorial lecture V, Oslo: Universitetsforlaget, 5-42). Taga has improved this idea, because “the rhythm generated in the central part is transmitted by sensory signals caused by the rhythmic movement of the motor organs…”, “motor output (motor output) It is an emergent property of dynamic interaction between the musculoskeletal system and the environment ”(Taga, G., 1995. A model of the neuro-musculo-skeletal system for human locomotion. I. Emergence of basic gait. Biol. Cybern. 73 (2), 97-111). Based on this, he combined CPG with sensory feedback and how the basic gait (gait) is from global entrainment between the nervous and musculoskeletal rhythmic activities. A neuromuscular model of human gait was demonstrated to demonstrate what happens.

  There is ongoing debate about what the actual ratio of central input and reflex input that generate motor output (motor output) means (Pearson, KG, 2004. Generating the walking gait: role Prog Brain Res 143, 123-129; Frigon, A., Rossignol, S., 2006.Experiments and models of sensorimotor interactions during locomotion. BiolCybern 95 (6), 607-627; Hultborn, H., 2006. Spinal reflexes, mechanisms and concepts: from Eccles to Lundberg and beyond.Prog Neurobiol 78 (3-5), 215-232; Prochazka, A., Yakovenko, S., 2007. The neuromechanical tuning hypothesis.ProgBrain Res 165, 255-265 ). For example, in walking cats, it has been estimated that only about 30% of the muscle activity (or muscle activity, the same applies hereinafter) observed with leg extensors that support weight can contribute to muscle reflexes ( Prochazka, A., Gritsenko, V., Yakovenko, S., 2002. Sensory controlof locomotion: reflexes versus higher-level control.Adv Exp Med Biol 508,357-367; Donelan, JM, McVea, DA, Pearson, KG, 2009. Force regulation of ankle extensor muscle activity in freely walking cats. J Neurophysiol 101 (1), 360-371).

  In humans, the contribution of reflexes (or reflexes) to muscle activity during locomotion appears to be more pronounced. Sinkjaer and colleagues estimated from reflex experiments that reflexes contribute approximately 50% to the activity of the soleus muscle during the stance phase of walking (Sinkjaer, T., Andersen, JB, Ladouceur, M , Christensen, LO, Nielsen, JB, 2000. Major role for sensory feedback in soleus EMG activity in the stance phase of walking in man. J Physiol 523 Pt 3, 817-827). Recently, Grey and colleagues found that the muscle activity of the soleus changes in proportion to changes in the tendon force of the Achilles tendon, and between positive force feedback (positive force feedback) and the action on this muscle. (Grey, MJ, Nielsen, JB, Mazzaro, N., Sinkjaer, T., 2007. Positive force feedback in human walking. J Physiol 581 (1), 99-105). It remains unclear whether such a large reflex nerve contribution exists in all leg muscles. As Daley and colleagues conclude from bird walking experiments, perhaps the proximal muscles of the lower limbs are controlled primarily by central input, while the distal muscles of the lower limbs are given greater proprioceptive feedback ( In situations where proprioceptive feedback is dominated by reflex input due to greater gain and sensitivity to mechanical action, there will be near-distal gradients in motion control (Daley, MA, Felix, G., Biewener, AA, 2007. Running stability isenhanced by a proximo-distal gradient in joint neuromechanical control. J ExpBiol 210 (Pt 3), 383-394).

  Application to topography (terrain) is an important aspect of walking. Commercially available short leg braces utilize a lightweight and passive structure designed to provide adequate elasticity during the stance phase of walking (S. Ron, Prosthetics and Orthotics: Lower Limb and Spinal. Lippincott Williams & Wilkins 2002). The advanced composite material used in these braces stores some energy during controlled dorsiflexion and plantar flexion, similar to the unaffected human Achilles tendon, and then drives It is possible to release energy during a strong plantar flexion (AL Hof, BA Geelen, Jw. Van den Berg, "Calf muscle moment, work and efficiency in level walking; role of series elasticity, "Journalof Biomechanics, Vol. 16, No. 7, pp. 523-537, 1983; DA Winter," Biomechanical motor pattern in normal walking, "Journal of Motor Behavior, Vol. 15, No. 4, pp. 302-330, 1983).

  This passive and elastic behavior closely resembles ankle function during slow walking, but normal walking and fast walking require additional external energy and hence normal and fast walking Cannot be performed with passive ankle braces (M. Palmer, "Sagittal plane characterization of normal human ankle function across a range of walking gait speeds," Master's Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2002; DH Gates, "Characterizing ankle function during stair ascent, descent, and levelwalking for ankle prosthesis and orthosis design," Master's Thesis, BostonUniversity, 2004; AH Hansen, DS Childress, SC Miff, SA Gard, KPMesplay, "The human ankle during walking: implication for the design of biomimetic ankle prosthesis, "Journal of Biomechanics, Vol. 37, Issue 10, pp. 1467-1474, 2004). This deficiency is reflected in the gait of leg amputees using passive ankle braces. Their self-selected walking speeds are slower than normal people and steps are shorter than normal people (DA Winter and SE Sienko. "Biomechanics of below-kneeamputee gait," Journal of Biomechanics, 21, pp. 361-367, 1988). In addition, their walking is clearly asymmetric. That is, the range of movement of the ankle joint on the unaffected side is smaller (HB Skinner and DJ Effeney, “Gait analysis inamputees,” Am J Phys Med, Vol. 64, pp. 82-89, 1985; H. Bateni and S. Olney, "Kinematic and kinetic variations of below-knee amputee gait," Journal of Prosthetics & Orthotics, Vol. 14, No. 1, pp. 2-13, 2002), while on the disabled side, Larger extension moments and smaller knee flexion moments (DA Winter and SE Sienko. "Biomechanics of below-kneeamputee gait," Journal of Biomechanics, 21, pp. 361-367, 1988; H. Bateniand S. Olney, "Kinematic and kinetic variations of below-knee amputeegait," Journal of Prosthetics & Orthotics, Vol. 14, No. 1, pp. 2-13, 2002). They also consume more metabolic energy in walking than those who have not cut their lower leg (NH Molen, "Energy / speed relation of below-knee amputeeswalking on motor-driven treadmill," Int. Z. Angew, Physio, Vol. 31, p 173, 1973; GR Colborne, S. Naumann, PE Longmuir, and D. Berbrayer, "Analysis of mechanical and metabolic factors in the gait of congenitalbelow knee amputees," Am. J. Phys. Med. Rehabil, Vol. 92, pp 272-278,1992; RL Waters, J. Perry, D. Antonelli, H. Hislop. "Energy cost ofwalking amputees: the influence of level of amputation," J Bone JointSurg. Am., Vol 58, No. 1, pp. 4246, 1976; EG Gonzalez, PJ Corcoran, and L. R. Rodolfo. Energy expenditure in B / K amputees: correlation with stumplength. Archs. Phys. Med. Rehabil. 55, 111-119 , 1974; DJ Sanderson and PEMartin. "Lower extremity kinematic and kinetic adaptations in unilateralbelow-knee amputees during walking," Gait and Posture. 6, 126 136, 1997; A. Esquenazi, and R. Di Giacomo. "Rehabilitation After Amputation," JournAm Podiatr Med Assoc, 91 (1): 13-22, 2001). These differences may be because amputees use more hip force to compensate for the lack of ankle strength (AD Kuo, "Energetics of actively powered locomotion using the simplest walking model," J Biomech Eng ., Vol. 124, pp. 113-120, 2002; AD Kuo, JM Donelan, and A. Ruina, "Energetic consequences of walkinglike an inverted pendulum: Step-sto-step transitions," Exerc. Sport Sci. Rev., Vol. 33, No. 2, pp. 88-97, 2005; A. Ruina, JE Bertram, and M. Srinivasan, "A collisional model of the energetic cost of support workqualitatively explains leg sequencing in walking and galloping, pseudo-elasticleg behavior in running and the walk-to-run transition. "J. Theor. Biol, Vol. 237, No. 2, pp. 170-192, 2005).

  Passive short leg braces cannot provide the ability to adapt to terrain. Currently, powered short-limb orthoses have been developed to provide normal, efficient walking beyond slow walking speeds (ie with an engine) (S. Au and H. Herr. "Initial experimental study on dynamic interaction between an amputee and a powered ankle-foot prosthesis, "Workshop on Dynamic Walking: Mechanics and Control of Human and Robot Locomotion, Ann Arbor, MI, May 2006; SK Au, J. Weber, and H. Herr," Biomechanicaldesign of a powered ankle-foot prosthesis, "Proc.IEEE Int. Conf.OnRehabilitation Robotics, Noordwijk, The Netherlands, pp. 298-303, June 2007; S.Au, J. Weber, E. Martinez- Villapando, and H Herr. "Powered Ankle-FootProsthesis for the Improvement of Amputee Ambulation," IEEE Engineering inMedicine and Biology International Conference. August 23-26, Lyon, France, pp. 3020-3026, 2007; H. Herr, J. Weber, S Au. "Powered Ankle-FootProsthesis," Biomechanics of the Lower Limb in Health, Disease and Rehabilitation. Septemb er 3-5, Manchester, England, pp. 72-74, 2007; SKAu, "Powered Ankle- Foot Prosthesis for the Improvement of Amputee WalkingEconomy," Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2007 ; S. Au, J. Weber, and H. Herr. "Powered Ankle-foot Prosthesis Improves Walking Metabolic Economy," IEEE Trans, on Robotics, Vol. 25, pp.51-66, 2009; J. Hitt, R. Bellman , M. Holgate, T. Sugar, and K. Hollander, "The sparky (spring ankle with regenerative kinetics) projects: Design andanalysis of a robotic transtibial prosthesis with regenerative kinetics," in Proc. IEEE Int. Conf. Robot. Autom. , Orlando, FL, pp 2939-2945, May 2006; SK Au, H. Herr, "On the Design of a Powered Ankle-Foot Prosthesis: TheImportance of Series and Parallel Elasticity," IEEE Robotics & Automation Magazine, pp. 52-59 , September 2008). Some of them have the same size and weight as an undisturbed human lower limb, allowing for storage of elastic energy, motive power (motor power), and batteries that allow typical walking activities of the day Equipped with energy (SK Au, H. Herr, "On the Design of a Powered Ankle-Foot Prosthesis: The Importance of Series and Parallel Elasticity," IEEE Robotics & Automation Magazine, pp. 52-59, September 2008).

  The use of active motive power (motor power) in these appliances creates control problems. Previous work on these powered braces has employed an approach to adapt the torque-ankle state profile of an unaffected human ankle to the activity being performed (SK Au, "Powered Ankle- Foot Prosthesis for the Improvement of Amputee Walking Economy, "Ph.D. Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2007; J. Hitt, R. Bellman, M. Holgate, T. Sugar, and K. Hollander," The sparky (spring ankle with regenerative kinetics) projects: Design and analysis of a robotic transtibial prosthesis with regenerativekinetics, "in Proc. IEEE Int. Conf. Robot. Autom., Orlando, FL, pp2939-2945, May 2006; F. Sup, A. Bohara, and M. Goldfarb, “Design and Control of a Powered Transfemoral Prosthesis,” The International Journal of Robotics Research, Vol. 27, No. 2, pp. 263-273, 2008). The supply of motive power (motor power) can support the open work loop of the angle-torque profile during faster walking, rather than just the spring-like behavior provided by passive appliances. . However, this control method does not provide inherent adaptability. On the contrary, a torque profile was required for all intended activities and terrain changes, and an appropriate means for selecting those profiles was also required.

  In general, existing commercially available active ankle joints can only reset the ankle angle during the swing phase and take several steps to converge to a terrain-matched ankle position upon first contact with the ground. Cost. Furthermore, they do not provide the stance driving force necessary for normal walking, and therefore the net work (amount) of stance cannot be adapted to sloping terrain. In particular, the powered short leg brace control scheme relies on a constant torque-ankle condition relationship obtained from measurements of an undisturbed human gait walking on a known terrain at a target speed. These controllers are effective at their intended walking speed and terrain, but do not allow them to adapt to environmental disturbances such as speed changes or terrain changes.

  In recent years, neuromuscular models that use positive force feedback reflex as the basis for control have been used in biomechanical simulation studies of leg movement (walking) (submitted for publication) “H. Geyer, H Herr, "A muscle-reflex model that encodesprinciples of legged mechanics predicts human walking dynamics and muscleactivities" and "H. Geyer, A. Seyfarth, R. Blickhan," Positive force feedback inbouncing gaits ?, "Proc. R Society. Lond. B 270, pp. 2173-2183, 2003 ". Such studies show promise for the need for adaptation to terrain.

  In one aspect, the present invention is a controller and control method for a biomimetic robot leg based on a neuromuscular model of human walking. The control architecture commands biomimetic torque to a powered prosthesis, orthosis, or exoskeleton ankle, knee and hip joints during walking. In a preferred embodiment, the powered device comprises artificial ankle and knee joints capable of torque control. The appropriate joint torque is determined by feedback information provided by a sensor mounted on each joint (each joint) of the robot leg device and provided to the user. These sensors include, but are not limited to, digital encoders that utilize angular displacement and angular velocity of joints, Hall effect sensors or the like, torque sensors at ankles and knees, and between knees and ankles. Includes at least one inertial measurement unit (IMU).

  Sensor information of the joint state (joint position and velocity) from the robot leg is used as an input to the neuromuscular model of human walking. The sensor information about the joint state from the robot leg is used to determine the internal state of each virtual muscle, and the level of specific muscle activity to be given to the strength and stiffness of each virtual muscle is determined by the spiral reflex (action). Determined from the model. When the robot leg is a prosthetic leg worn by a person who cuts the thigh, the joint state of these joints is measured by angle sensors at the ankle joint (ankle) and knee (joint). For the hip joint, the absolute orientation (or absolute orientation) of the user's thigh is determined using an angle joint sensor in the artificial knee and an IMU placed between the joint of the artificial knee and the ankle joint. In order to estimate the position and velocity of the hip joint, the control architecture operates under the assumption that the upper body (torso) maintains a relatively vertical posture during walking.

  In one aspect, the invention is a model-based neuromechanical controller for a robotic limb (robot hand and / or foot) having at least one joint, the controller receiving feedback data regarding the state of the robotic limb. A finite state machine configured to determine the state of the robotic limb, state information from the finite state machine, and muscle geometry / arrangement and reflex architecture information from at least one database A muscle model processor configured to receive and use the neuromuscular model to determine at least one desired joint torque command or stiffness command to be sent to the robot limb, and determined by the muscle model processor at the joint of the robot limb Command to provide biomimetic torque and stiffness Comprising a joint instruction processor configured. In a preferred embodiment, the feedback data is provided by at least one sensor mounted on each joint of the robot limb. In another preferred embodiment, the robot limb is a leg and the finite state machine is synchronized to the leg walking cycle.

  In another aspect, the invention is a model-based method for controlling a robotic limb that includes at least one joint, and receiving feedback data (also referred to as feedback data) regarding the state of the robotic limb in a finite state machine. Determining the state of the robot limb using the finite state machine and the received feedback data; a neuromuscular model, muscle geometry / placement information, and reflex architecture information; and Determining at least one desired joint torque command or stiffness command to be sent to the robot limb using state information from the finite state machine; and a biomimetic torque determined by the muscle model processor at the joint of the robot limb And commanding stiffness to occur.

  Other aspects, advantages and novel features of the invention will become more apparent from the following detailed description of the invention and the accompanying drawings.

1 is a block diagram of an exemplary embodiment of a general neuromuscular model architecture according to one aspect of the present invention. FIG. A through F illustrate each of the six stages of development of a general neuromuscular model architecture in accordance with one aspect of the present invention. Figure 2 is a visual representation of pattern generation according to one aspect of a general neuromuscular model architecture according to the present invention. A shows the walking of a self-organized human body model from the dynamic interaction between the model and the ground according to one aspect of the present invention, and B is the corresponding floor reaction force. (Also called ground reaction force). In A, B, and C, according to one aspect of the present invention, steady walking (walking in a steady state) between the model and the subject with respect to the hip joint (or hip), knee (joint), and ankle joint (or ankle), respectively. Is shown in comparison. A, B, C, and D, respectively, are model snapshots, net work, extensor activity patterns, and corresponding floor reaction forces according to one aspect of the present invention. Is shown. A, B, C, and D are model snapshots, net work, extensor activity patterns, and corresponding floor reaction forces, respectively, according to one aspect of the present invention. Is shown. 1 is a schematic illustration of a muscle-tendon model according to one aspect of the present invention. 2 illustrates a contact model according to one aspect of the present invention. 1 illustrates an exemplary embodiment of an ankle foot orthosis used in a preferred embodiment according to one aspect of the present invention, wherein A, B, and C are the physical system, (drive train (power transmission system), respectively) Shows the machine model. FIG. 4 illustrates an exemplary embodiment of a finite state machine synchronized to a walking cycle in accordance with one aspect of the present invention, the finite state machine including a state transition threshold and an equivalent ankle-foot in each state. And are used to implement the highest level of control of the ankle foot orthosis of FIGS. 10A-10C. 1 is a block diagram of an exemplary embodiment of a control system for a short leg brace according to one aspect of the present invention. FIG. In accordance with one aspect of the present invention, a plot of orthosis torque over a complete walking cycle for three walking conditions: horizontal ground (shown in A), uphill (shown in B), downhill (shown in C). It is an example. FIG. 4 illustrates an exemplary embodiment of a musculoskeletal model implemented in a brace microcontroller according to one aspect of the present invention, where A is a two-link ankle model and B is a detailed hill-type muscle model ( Hill-type muscle model), C, shows the geometry / arrangement of muscle model skeletal attachment. 6 illustrates an exemplary embodiment of a virtual plantar flexor reflex scheme, including a relationship between ankle angle, muscle strength, and plantar flexor component of ankle torque, in accordance with an aspect of the present invention. A and B are respectively measured during the experiment between the brace worn by the subject who lost the limb (especially the lower limb) and the biological ankle of the subject whose weight and height are not damaged by the lower limb. FIG. 5 is a diagram showing a comparison of trajectories of torque (including ankle joint torque) and angle (including ankle joint angle). FIG. 4 shows a torque profile after optimizing parameters compared to a biological torque profile according to one aspect of the present invention. One exemplary embodiment of the present invention was measured experimentally on three different walking conditions: horizontal ground (shown in A), uphill (shown in B), downhill (shown in C). It is the figure which plotted the locus | trajectory of the torque-angle of an orthosis.

  A control architecture is provided to provide biomimetic torque commands to the powered prosthetic leg, orthosis, or exoskeleton ankle, knee and hip joints during walking. In this embodiment, the powered device comprises artificial ankle and knee joints that can control torque. The appropriate joint torque determined by the feedback information provided by the sensor mounted on each joint of the robot leg device is provided to the user. These sensors include, but are not limited to, digital encoders that utilize angular displacement and angular velocity of joints, Hall effect sensors or the like, torque sensors at ankles and knees, and between knees and ankles. Includes at least one inertial measurement unit (IMU).

  Sensor information of joint states (joint position and velocity) from robot legs (hip joints, knee joints, and ankle joints) is used as input to a neuromuscular model of human walking. This model uses the sensor information about the joint state from the robot leg to determine the internal state of each virtual muscle, and to determine the strength and stiffness of each virtual muscle determined from the spiral reflex (action) model. Establish the level of specific muscle activity to be given. When the robot leg is a prosthetic leg worn by a person who cuts the thigh, the joint state of these joints is measured by angle sensors at the ankle joint and knee joint. For the hip joint, the absolute orientation (or absolute orientation) of the user's thigh is determined using an angle joint sensor in the artificial knee and an IMU placed between the joint of the artificial knee and the ankle joint. In order to estimate the position and velocity of the hip joint, the control architecture operates under the assumption that the upper body (torso) maintains a relatively vertical posture during walking.

  The meanings of some terms used in this specification, as well as in applications incorporated herein by reference, are set forth below, but are not limited to those meanings.

  “Actuator” means a type of motor defined as described below.

  “Agonist muscle” means a contractile element that is subjected to resistance or reaction by another element, an antagonist muscle.

  “Actuator-antagonist actuators” are (at least) two actuators that act oppositely to each other: an actuator actuator that brings two elements closer when energized and a distance between the two elements when energized It means a mechanism (mechanism) composed of an antagonistic muscle actuator.

  “Antagonist muscle” means an inflatable element that is subjected to resistance or reaction by another element, the agonist muscle.

  “Biomimetic” means an artificial structure or mechanism that mimics (imitates) the characteristics and behavior of biological structures such as joints and limbs and biological mechanisms.

  “Dorsal flexion” means bending the ankle joint so that the tip of the foot moves upward.

  "Elasticity" means the ability to recover to its original shape after deformation due to stretching or compression.

  “Extension” means a bending motion around the joint of the limb that increases the angle between the bones of the limb at the joint.

  “Bend” means a bending motion around the joint of the limb that reduces the angle between the bones of the limb at the joint.

  “Motor” means an active element that creates or imparts motion by converting supplied energy into mechanical energy (or mechanical energy), and includes an electric motor / electric actuator, a pneumatic motor / pneumatic actuator, Or it includes a hydraulic motor / hydraulic actuator.

  “Planar flexion” means bending the ankle joint so that the tip of the foot moves downward.

  “Spring” means an elastic device such as a metal coil or leaf structure that returns to its original shape after compression or extension.

  One exemplary embodiment of a neuromuscular model-based control scheme according to this aspect of the invention is shown in the block diagram of FIG. In FIG. 1, a neuromuscular model according to the present invention has a reflection loop (reflection architecture) 110 for each modeled muscle unit 120. Predicted forces and stiffness from all modeled muscles are used to calculate (130) joint torque and stiffness using the muscle moment arm values 140 obtained from the literature. The calculated values of this model are sent to the controller as the desired net torque and stiffness values for the biomimetic robot leg joint 150. The controller 160 tracks the torque value and stiffness value at each of the robot joints 150.

  In order for each of the virtual muscles to produce the required force, a muscle stimulation (muscle stimulation) parameter STIM (t) is required. This parameter can be determined from an external input or a local feedback loop. In an exemplary biomimetic leg control method, STIM (t) is calculated based on a local feedback loop. This architecture is based on the reflection feedback framework developed by Geyer and Herr ("H. Geyer, H. Herr," A muscle-reflex model that encodesprinciples of legged mechanics predicts human walking dynamics and muscleactivities "). The entire contents of the article are incorporated herein by reference. In this framework, neural control is designed to mimic the stretch reflex of an unhindered (ie undamaged) human muscle. This control method based on neuromuscular reflex enables a biomimetic robot leg to replicate joint mechanics (or mechanism) similar to humans.

  Neural machine model. A human body model with reflex control that encodes the principles of gait dynamics (gait mechanism) predicts human gait dynamics and muscle activity. While neuroscientists are gradually finding complex neural networks that control the walking of animals and humans, biomechanics researchers can focus on the principles of gait mechanics, We find that little control is needed. Here we show how important muscle reflex (also called muscle reflex) behavior can be to relate these two views. The developed human walking model is driven by muscle reflex behavior that encodes the principles of walking mechanics. By providing reflection control based on this principle, the model stabilizes from dynamic interaction with the ground to a walking state, withstands ground fluctuations, and self-adapts to stairs. In addition, the model is qualitatively consistent with experimentally known joint angles, joint torques, and muscle activity, and the human motor output is a muscle reflex that associates the principles of walking mechanics with the neural network responsible for walking. It suggests that it can be formed mainly by activities.

  Human gait models with motor control are based on muscle reflexes designed to include such gait dynamics principles. These principles are derived from a simple conceptual model of leg walking. Blickhan, R., 1989. The spring-mass model for running and hopping. J. of Biomech. 22, 1217- 1227; Ghigliazza, R., Altendorfer, R., Holmes, P., Koditschek, D., 2003.A simply stabilized running model.SIAM J.Applied.Dynamical Systems 2 (2), 187-218; Geyer, H. , Seyfarth, A., Blickhan, R., 2006. Compliant leg behavior explains the basic dynamics of walking and running. Proc. R. Soc. Lond. B 273, 2861-2867), segmentation based on static joint torque balance Stabilized Legs (Seyfarth, A., Gunther, M., Blickhan, R., 2001. Stable operation of an elastic three-segmented leg. Biol. Cybern. 84, 365-382; Gunther, M., Keppler, V., Seyfarth, A., Blickhan, R., 2004. Human leg design: optimal axialalignment under constraints. J. Math. Biol. 48, 623-646), ballistic swing-legmechanics (Mochon, S., McMahon, T., 1980. Ballistic walking. J. Biomech. 13 (1), 49-57) and improved walking stability using swing-leg retraction (Seyfarth, A., Geyer, H., Gunther, M., Blickhan, R., 2002. A movementcriterion for running.J. Of Biomech.35, 649-655; Seyfarth, A., Geyer, H., Herr, HM, 2003.Swing-leg retraction: a simple control model for stablerunning. J. Exp. Biol. 206, 2547-2555). Using hill-type muscles with positive force and length (or distance) feedback combined with spiral reflexes, they effectively encode their mechanical properties.

  Designed to encode the principles of gait mechanics when comparing the behavior of this model with kinetic, kinematic and electromyographic evidence from the human gait literature. It has been found that a neuromuscular model with motor control can generate biological gait mechanisms and muscle activity. With this reflection control, the model can withstand sudden changes in ground height without parameter intervention and can adapt to ascending and descending stairs.

  The structure and control of the human body model goes from a conceptual point-mass model through six stages to the upper body and two or three, each controlled by muscle reflexes, moved by seven muscles. Progresses to bipedal animals with neuromuscular activity with segmented legs. 2A-F illustrate the six stages of development of a general neuromuscular model architecture in accordance with this aspect of the invention. In the first three stages (A to C in FIG. 2), the compliant leg movements in the standing state (or adapted to the standing state) are integrated and stabilized. In the fourth stage (D in FIG. 2), the upper body and its balance control are added. In the last two stages (E and F in FIG. 2), preparation for leg protrusion and retraction during the swing phase is made and ensured.

  2A-F described in more detail in the following paragraphs, starting with the leg configuration during standing (FIG. 2A), the soleus muscle (SOL) and the concentrated vastus muscles with positive force feedback F + By driving (vasti group muscles: VAS), a compliant leg movement is generated as a key for walking and running (an important means) (B in FIG. 2). In order to prevent overextension of the knee, the gastrocnemius muscle (GAS), a biarticular muscle using F +, was added (C in Fig. 2), and VAS was suppressed when the knee joint extended beyond the threshold of 170 ° Is done. An anterior tibial muscle (TA) is added to prevent ankle overextension, and retracting the ankle into position with positive feedback L + of the length by the anterior tibial muscle under normal stance Is suppressed by negative force feedback F- from the soleus. The upper body is added to allow swinging of the foot (D in FIG. 2). The upper body is tilted by a reference angle with respect to the vertical direction by the hip flexor (HFL) of the foot and the hip extension muscle (GLU, HAM) driven together, and in this case by the HAM of the biarticular muscle Overextension of the knee resulting from the hip extension muscle moment is prevented. By adding / removing certain stimuli to / from the HFL / GLU respectively and suppressing the VAS in proportion to the load on the other leg, the other (predecessor) leg in the landing state Shaking is started (E in FIG. 2). The actual leg swing is promoted by the HFL using L + until L + is suppressed by L− of HAM (F in FIG. 2). HFL stimuli are biased depending on the inclination of the upper body as it leaves the ground. In addition, by using F + for GLU and HAM, the leg is retracted and straightened when the swing is near the end. Finally, the ankle is moved to a fixed position by the now unrestrained L + of TA (G in FIG. 2).

  Leg compliance and stability when standing. A bipedal spring-mass model is used as a starting point for the conceptual basis of human walking (FIG. 2A). This model is composed only of the mass point 205 that travels on the two massless spring legs 210, 220, but reproduces the mass center dynamics observed during human walking and running, and is compliant when standing. Based on leg movement, both those walking and running are integrated into one conceptual framework (Geyer, H., Seyfarth, A., Blickhan, R., 2006. Compliant leg behaviors explained the basic dynamics of walking and running. Proc. R. Soc. Lond. B 273,2861-2867).

  To achieve compliant behavior of the neuromuscular leg, each spring 210, 215 is replaced with a thigh 220, a shin 225, and a foot 230, and a soleus (SOL) 235 and a broad muscle group (VAS) 240 are added. The During the stance phase of gait, these muscles perform muscle activity with local positive force feedback (F +) (B in FIG. 2). This force reflection is described in Geyer, H., Seyfarth, A., Blickhan, R., 2003, “Positive force feedback in bouncing gaits? Proc. R. Soc. Lond. B 270, 2173-2183”. Modeled in the same way as In the state of positive force feedback, the stimulation Sm (t) of the muscle m is the sum of the preliminary stimulation S0, m and the muscle force Fm delayed by (Δt) and multiplied by the gain (G). That is, Sm (t) = S0, m + GmFm (t−Δtm).

Obedient leg movement is extremely important, but also threatens the stability of the segmented leg joint (Seyfarth, A., Gunther, M., Blickhan, R., 2001. Stable operation of an elastic three-segmented leg Biol. Cybern. 84, 365-382; Gunther, M., Keppler, V., Seyfarth, A., Blickhan, R., 2004. Human leg design: optimal axial alignmentunder constraints. J. Math. Biol. 48, 623-646). In the segmented leg, the knee joint torque τ _k and the ankle joint torque τ _a follow _a static equilibrium τ _k / τ _a = h _k / ha. Here, h _k and h _a are vertical distances from the knee joint and the ankle joint to the leg force vector Fleg, respectively. In fact, a large extension torque in one joint causes the other joint to be brought closer to the Fleg, causing the leg that behaves like a spring to overextend (for details see Seyfarth, A., Gunther, M ., Blickhan, R., 2001. See Stable operation of an elastic three-segmented leg. Biol. Cybern. 84, 365-382).

This tendency to overextend at the knee or ankle joints is caused by the addition of gastrocnemius (GAS) 245 and anterior tibial muscle (TA) 250 (FIG. 2C). Similar to SOL and VAS, biarticular GAS uses local positive force feedback (F +) during the stance phase of gait. This muscle reflex not only prevents overextension of the knee joint due to a large extension torque at the ankle joint, but also contributes to the realization of the compliant overall leg behavior. In contrast, single-joint TA uses local positive length feedback (L +), where S _TA (t) = S _{0, TA} + G _TA (l _{CE, TA} −l _{off , TA} ) (t−Δ _{t, TA} ). Here, l _{CE and TA} are TA fiber lengths, and l _{off and TA} are length offsets. By bending the foot, TA's L + prevents the ankle joint from overextending when large knee joint torque is generated. However, this muscle reflex is not necessary if the knee and ankle torque balance is maintained by sufficient activity of the ankle extensors. To avoid unnecessarily counteracting SOL in this situation, TA stimulation is suppressed by negative force feedback (F-) from SOL, resulting in S _TA (t) = S _{0, TA} + G _TA (l _{CE, TA} -l _{off, TA} ) (t- _{Δt, TA} ) -G _SOLTA F _SOL (t-Δt _SOL ) Further, in order to prevent the knee joint from overextending, when the knee joint extends beyond the threshold value of 170 °, VAS is suppressed and S _VAS (t) = S _{0, VAS} + G _VAS F _VAS (t−Δt _VAS ) −k _ψ Δψ _k (t−Δt _k ). Here, k _ψ is a proportional gain (proportional gain), Δψ _k = ψ _k −170 °, and ψ _k is a knee joint angle. This reflection suppression only works when ψ _k > 0 and the knee joint is actually extended.

The upper body and its balance. In the next step, which evolves from a conceptual spring-mass model to a bipod with neuromuscular motion, the mass representation is discarded and the upper body (upper body) 255 (can swing the leg around the upper body) Is introduced (D in FIG. 2). The upper body 255 is coupled to the head, arms, and torso (these are abbreviated as HAT). In order to balance the HAT 255 during walking, a gluteal muscle group (GLU) 260 and a hip flexor group (HFL) 265 are added to each of the legs. GLU260 and HFL265 are proportional-derivative signals of forward tilt angle (tilt angle forward) θ with respect to the direction of gravity of HAT255, S _{GLU / HFL˜} ± [k _p (θ−θ _ref ) + k _d dθ / Dt]. Here, k _p is a proportional gain (proportional gain), k _d is a differential gain (differential gain), and θ _ref is a reference tilt angle (for a similar approach, see, for example, “Gunther, M., Ruder, H., 2003. Synthesis of two-dimensional human walking: a test of the λ-model. Biol. Cybern. 89, 89-106 ”). In order to combat the knee hyperextension caused by the large hip torque generated by GLU 260 when lifting heavy HAT 255, the bi-articular hamstring muscle group (HAM) 270 with S _{HAM to} S _GLU is also included. include. If the legs support a sufficient weight, only the hip joint torque can balance (maintain equilibrium) with the HAT 255, so the stimulation of the GLU 260, HAM 270 and HFL 265 is proportional to the weight supported by each leg. Adjusted for each leg. As a result, the hip joint muscles of each leg contribute to HAT balance control only during the stance phase.

である。この対側性の抑制によって、膝関節は、ばねのように機能する動作を中断し、及び、足関節が伸びている間に曲がって、脚を地面から離して前方に進めることが可能になる。このカタパルト（または射出機）機構は、足関節が十分に押し返す場合にのみスイング動作を開始することが可能であるが、該モデルは、さらに、両脚支持期において一定量ΔSだけ、HFL２６５の刺激を大きくし、GLUの刺激を小さくすることによって脚を繰り出すスイング動作の準備をする。 Swinging out the legs (protruding) and retracting the legs (leg swinging). The structure of the human body model is completed except for the muscle reflex control that generates a swing motion that extends the leg and retracts the leg. The importance of leg function in the stance phase is reduced in proportion to the weight (bw) supported by the leg on one side of the pair of legs (ie, the opposite side), and the two-leg support period In FIG. 2, it is assumed that the swing motion of extending the leg is started as early as (E in FIG. 2). The human body model detects which leg last entered the stance phase (which leg is the opposite leg of the pair of legs) and is proportional to the weight supported by the opposite leg. The F + of the VAS 240 on the other leg (ie, the same side) is suppressed. in this case,

It is. This contralateral restraint causes the knee joint to cease to act like a spring and bend while the ankle is extended, allowing the leg to move forward away from the ground. . Although this catapult (or injection machine) mechanism can only start swinging when the ankle joint pushes back sufficiently, the model further stimulates HFL265 by a certain amount ΔS in both leg support periods. Prepare for swinging action to extend the leg by increasing the size and reducing the stimulation of GLU.

During the actual swing motion, we rely primarily on the ballistic motion of the legs, which is affected in two ways (F in FIG. 2). On the other hand, the swinging leg can be easily extended. The HFL 265 is stimulated with positive feedback (L +) of length biased by the forward pitch angle θ _ref of the HAT 255 during the transition from the stance phase to the swing phase. The stimulus is S _HFL (t) = S _{0, HFL} + k _lean (θ−θ _ref ) _TO + G _HFL (l _{CE, HFL} −l _{off, HFL} ) (t−Δ _{t, HFL} ). By using this approach, it is ensured that the ballistic movement of the swinging leg gains momentum to advance the leg forward without time delay (Mochon, S., McMahon, T., 1980 Ballistic walking. J. Biomech. 13 (1), 49-57).

Further, the swinging leg is prevented from being excessively stretched, and the leg is reliably retracted. During the swing phase, if the leg stretches and maintains the proper orientation, the walking system self-stabilizes and enters the walking cycle (McGeer, T., 1990. Passive dynamic walking. Int. J. Rob. Res. 9 (2 ), 62-82; Seyfarth, A., Geyer, H., Gunther, M., Blickhan, R., 2002. A movementcriterion for running. J. of Biomech. 35, 649-655; Ghigliazza, R., Altendorfer , R., Holmes, P., Koditschek, D., 2003.A simply stabilized running model.SIAM J.Applied.Dynamical Systems 2 (2), 187-218; Geyer, H., Seyfarth, A., Blickhan, R., 2006. Compliant leg behavior explains the basic dynamics of walking and running. Proc. R. Soc. Lond. B 273, 2861-2867). The resistance of this mechanical self-stability to disturbance can be greatly improved if the swinging leg is additionally retracted before landing on the ground (Seyfarth, A., Geyer, H., 2002. Natural control of spring-likerunning-optimized self-stabilization.In: Proceedings of the 5thinternational conference on climbing and walking robots.Professional Engineering Publishing Limited, pp. 81-85; Seyfarth, A., Geyer, H., Herr, HM, 2003.Swing- leg retraction: a simple control model for stable running. J. Exp. Biol. 206, 2547-2555). In order to implement this “stop-withdrawal” scheme, three muscle reflexes are provided in the human body model. Excessive stretching (overextension) of a swinging leg is caused by a forward impulse received by the leg when the knee joint extends to a fully extended state while the leg is extended. Overpass is prevented. Here, L + of HFL is suppressed in proportion to the elongation that HAM receives during the swing operation. In this case, S _HFL (t) = k _lean (θ−θ _ref ) TO + G _HFL (l _{CE, HFL} −l _{off, HFL} ) (t−Δ _{t, HFL} ) −G _HAMHFL (l _{CE, HAM} −l _{off, HAM} ) (t− _{Δt, HAM} ). Further, F + is used for GLU and HAM, respectively, S _GLU (t) = S _{0, GLU} + G _GLU F _GLU (t−Δt _GLU ), S _HAM (t) = S _{0, HAM} + G _HAM F _HAM (t−Δt _HAM ), so that depending on the actual payout momentum, the swinging leg not only stops, but also a part of this momentum is used to straighten the leg and Is used (converted) for entrainment. Finally, the TA L + introduced to ensure foot clearance (or foot-to-foot distance) is maintained throughout the swing motion. SOL, GAS and VAS are inactive during this swing phase.

Reflection control parameter. The contribution of each reflex effect to the muscle stimulation Sm (t) is determined by the formula used in the model. No parameter optimization was performed. The parameters were obtained from previous knowledge of reflex motion (F +, L +) or were obtained by making plausible estimates. All muscle stimulation is limited to the range of 0.01 to 1 before being converted to muscle activity _Am (t). Table 1 shows the stance phase reflection equation used in the preferred embodiment.

Table 2 shows the swing equation during the swing phase used in the preferred embodiment.

result. The human body model does not have a central pattern generator (CPG) that activates its muscles by feedforward control, but it uses a bulge and heel at the base of each thumb to detect the ground. A sensor is used to switch different reflexes (or reflexes) for standing leg and swing leg for each foot. As a result, the dynamic interaction between the model and its mechanical environment is an indispensable element in the generation of muscle activity. FIG. 3 visually illustrates pattern generation according to this aspect of the invention. In FIG. 3, instead of the central pattern, reflexes generate muscle stimulation Sm 305, 310. The left (L) 320 leg and the right (R) 330 leg have separate standing leg reflexes 340 and 345 and free leg reflexes 350 and 355, which are connected to the foot. Selected based on contact detection 360, 365 from sensors 370, 375 at the bulge and heel of the thumb. The output of reflexes depends on mechanical inputs M _i 380, 385, intertwining mechanics, and motor control.

Walking state. To examine how important this interdependence of mechanism and motor control is to human walking, the model was started from the left leg during the stance phase and from the right leg during the swing phase, It was moved at a walking speed v0 = 1.3 m / sec. Since this modeled muscle reflex time delay is a maximum of 20 ms, initially all muscles are inactive. FIG. 4A illustrates the walking of a human body model that is self-organized from the dynamic interaction between the model and the ground in accordance with one aspect of the present invention; The corresponding floor reaction force (GRF) is shown. 4A shows a snapshot of a human body model taken every 250 ms, and FIG. 4B shows a corresponding model GRF. The left leg 405, 410 and the right leg 415 are shown. , 420 are plotted separately (with 30 Hz low pass filtering). The model starts walking at a horizontal speed of 1.3 m / s and decreases the speed after walking the first two steps, but rapidly recovers to the same walking speed. Only the leg muscles of the right leg 415 showing muscle activity> 10% are shown. The initial conditions for each leg ψ _{a, k, h} (definition of ankle, knee and hip joint angles) are (for the left leg) ψ _{a, k, h} = 85 °, 175 °, 175 °, (Regarding the right leg) ψ _{a, k, h} = 90 °, 175 °, 140 °.

  Because of these disturbed initial conditions, the model collapses slightly and slows down in the first step (A in FIG. 4). However, if the parameters are properly selected, the model recovers rapidly during subsequent walking and self-organizes the walking from the dynamic interaction between the model and the ground. Here, the floor reaction force (GRF) in the vertical direction of the leg in the stance phase shows an M-shaped pattern characteristic in the walking state (B in FIG. 4), which is related to the whole body between the model and the human with respect to the steady walking. This suggests similarities in dynamics (whole body dynamics).

  Steady state pattern of angle, torque and muscle activity. This similarity also withstands closer inspection. That is, a known angle, torque, and muscle activity pattern from human gait data are quantitatively matched with those of the model. FIGS. 5A, 5B, and 5C show models for hip joints (FIG. 5A), knee joints (FIG. 5B), and ankle joints (FIG. 5C), according to one aspect of the present invention. It is the figure which compared each subject's normal walk in 1.3m / sec. In FIGS. 5A to 5C, the steady state pattern of the muscle activity, torque, hip joint angle, knee joint angle, and ankle joint angle of the model is one stride from the heel contact to the heel contact of the same leg. Normalized to (step length) and compared with human walking data (Source: Perry, 1992). A vertical dotted line 510 in the vicinity of 60% of the stride indicates that the toes are separated (toe separation). The muscles to be compared are the long adductor muscle (HFL) 520, the upper gluteus maximus (GLU) 530, the semi-membranous muscle (HAM) 540, and the lateral vastus muscle (VAS) 550.

The best match between model prediction and gait data can be found at the ankle joint (C in FIG. 5). The reflex model not only generates the ankle motion ψ _a and torque τ _a observed for the human ankle during walking, but also for SOL, TA, and GAS activities based on experiments inferred from surface electromyography. Predict similar SOL, TA and GAS activities. For SOL and GAS, this activity is generated exclusively by their local F + reflex in the stance phase. For TA, its L + reflex responds with higher activity to plantar flexion in the early stance phase, but is suppressed by F- from the SOL for the remainder of the stance phase. Only when the SOL activity becomes smaller during the transition from the stance phase to the swing phase (at 60% of the stride), the L + of the TA starts to pull the foot to perform plantar flexion.

From this comparison, it can be seen that the degree of matching in the case of the knee joint and the hip joint is smaller. For example, a general trajectory ψ _{k of} a human knee is obtained by the model, but in the early stance phase, the knee joint of the model bends about 10 ° or 30% more than that of a human (see FIG. 5 B). In connection with this larger knee flexion, in the model no VAS activity is observed in the late (or end) swing phase following the early stance phase. Only after heel contact (phase) is the VAS F + involved, and the F + can activate (activate) muscle groups in response to the load (weight) on the leg. Due to the delay in extensor activity, not only the relatively weak knee (joint) in the early stance phase, but also the thick (or heavy) HAT leans forward after a collision (to the ground). Since balance control of the HAT is gradually involved in the weight (weight) supported by the leg in the stance phase, balance reflexes are not active until heel contact, Abnormally large GLU activity and HAM activity must be generated to return the HAT to the reference tilt level (reference tilt angle) (FIG. 5C). Thus, the hip trajectory ψ _h and torque pattern τ _h of the model are human (in the case of humans, the hip extension muscles GLU and HAM are in an active state before the collision and the torso tilts so much Is not most similar to the locus and torque pattern.

  Self-adaptation to ground changes: The human body model withstands sudden changes in ground height despite the limited reflex control and self-adapts to permanent ground changes. FIGS. 6A-6D show examples where the model encounters a series of stairs where each step is 4 cm higher. FIG. 6 illustrates the adaptation to walking up the stairs in accordance with one aspect of the present invention, with a snapshot of the model (A in FIG. 6), net work (B in FIG. 6). , The extensor muscle activity pattern (C in FIG. 6) and the corresponding floor reaction force (D in FIG. 6). 6A to 6D show eight strides of a human body model starting from steady walking at 1.3 m / second, and each of the five steps is tilted by 4 cm. The model returns to steady walking on the 8th stride. One stride refers to the right foot from heel contact to heel contact. FIG. 6A shows a snapshot of the model when the heel of the right leg is touched (when the heel touches the ground) and when the right foot is released from the toe (when the toe leaves the ground). . For this leg, FIG. 6B shows the net work that occurs at the hip, knee, and ankle joints during the stance phase (positive work is the work during extension (extension)), and FIG. The activity pattern of the five extensors in each stride is shown in FIG. 6D, where the corresponding floor reaction force 650 normalized to body weight (bw) is the floor reaction force 660 of the left leg for comparison. Along with.

  The model starts with a steady walk (first stride) and at the end of the second stride, the fully extended right leg foot reaches the stairs (A in FIG. 6). This early collision slows down the model, causing the upper body to lean forward and the upper body to receive a large hip torque generated by GLU and HAM (the third stride of FIGS. 6B and 6C). ). Hip extension torque also tends to extend the knee (knee joint), so even if the net work (amount) at the knee joint during the stance phase is maintained at the same level as at steady state, VAS It does not receive the same level of force as the time, and its muscle stimulation is reduced by the force feedback control (C in FIG. 6). In contrast, the deceleration of the model reduces the forces experienced by the ankle extensors GAS and SOL during the stance phase, and their force feedback reflexes produce slightly smaller muscle stimuli, The net work (amount) is reduced (B and C in FIG. 6). In strides 4 and 5, the model settles into a walking state that climbs the stairs at about 1 m / sec, where forward and upward thrust is generated primarily at the hip and knee joints. The model returns to the original steady walking speed of 1.3 m / sec in the eighth stride after reaching the stable period in the sixth stride.

  7A-D illustrate a model snapshot (FIG. 7A), net work (FIG. 7B), extensor activity pattern (FIG. 7C), and in accordance with one aspect of the present invention. , Showing adaptation to walking down the stairs with the corresponding floor reaction force (D in FIG. 7). In FIGS. 7A to 7D, eight strides are shown starting from a steady walking at a speed of 1.3 m / second of the human body model, and each of the five steps is tilted by 4 cm. The model returns to steady walking on the 8th stride. One stride is from heel contact to right heel contact. FIG. 7A shows a snapshot of the model at right leg heel contact and toe off. For this leg, FIG. 7B shows the net work generated at the hip, knee and ankle joints during the stance phase (the positive work is the work during extension (extension)). The activity pattern of the five extensors of each stride is shown in FIG. 7D, where the corresponding floor reaction force 750 normalized for body weight (bw) is compared with the floor reaction force 760 for the left leg for comparison. Further shown. The model goes back to a steady walk of 1.3 m / sec on the 14th stride after taking 5 steps of walking, each tilted by 4 cm.

  7A to 7D show a walking sequence when the model encounters a downstairs. The model is descending the first stair with the right foot at the end of the ninth stride (A in FIG. 7). This downward movement accelerates the model, and in the 10th stride, the right leg has an overall greater initial impact, with most extensors responding more strongly (C and D in FIG. 7). ). It is only GAS that produces less force, because with this stride, the knee joint remains more flexed than usual. As a result, the net positive work at the ankle joint is greatly increased (B in FIG. 7). This increase, as well as the greater HFL stimulation (not shown) caused by the forward tilt of the upper body when the foot leaves the ground, increases the stride during the swing phase and propels the right leg forward (FIG. 7). A). The model maintains this greater stride during the downward movement after the transient 10th stride (11th and 12th strides), and the downward acceleration of the model is (ground Counteracted by increased activity of GLU, HAM and VAS immediately after the collision (C and D in FIG. 7), which reduces net positive work at the hip joint and net negative work at the knee joint. Increases (B in FIG. 7) and stabilizes the model to a walking state of about 1.5 m / sec. When there is no further downward acceleration after returning to the flat ground, the model decelerates, thereby automatically reducing the stride (A in FIG. 7) and 1.3 m / sec on the 13th to 14th walks. It is driven to return to steady walking.

  A single control is not sufficient for both up and down stairs. What is important for the tolerance and adaptability of the model is its dynamic muscle reflex response. The bounce of the leg in the stance phase depends on the amount of load that the leg extensors SOL, GAS and VAS are subjected to, and this bounce ensures that the leg bends sufficiently so that it can move forward when going up the stairs. And it is guaranteed that the foot will brake sufficiently when going down the stairs. On the other hand, propulsion of the swing leg forward depends on the dynamics of the model. The abrupt deceleration after the collision of the opposite leg (to the ground) near the end of the stance phase, tilting the upper body forward, and the rate of extension of the ankle (or extension speed) are all during the swing phase. Contributes to leg propulsion. These combined features ensure that the free leg is fully extended when walking up the stairs and fully extended when walking down the stairs. In the latter case, GLU and HAM force feedback suppresses excessive rotation of the leg and allows the leg to be retracted and extended quickly.

Muscle tendon unit. All 14 muscle-tendon units (MTUs) of a bipod have the same model structure. FIG. 8 is a schematic illustration of a muscle-tendon model according to one aspect of the present invention. In FIG. 8, the working contractile element (CE) 810 forms a muscle-tendon unit (MTU) during normal operation with the elastic bodies (SE) 820 connected in series. When _CE 810 extends longer than its optimal length l _CE 830 (l _CE > l _opt 840), an elastic body (PE) 850 connected in parallel is involved. Conversely, the buffer elastic body (BE) 860 prevents the working _CE 810 from collapsing when the SE 820 is slack (l _MTU 870-l _CE 830 <l _slack 880).

As shown in FIG. 8, an active hill type contractile element (CE) produces a force along a series of elastic bodies (SE). The MTU is fitted into the skeleton so that individual CEs mainly act on the ascending limb in their force-length relationship, but the MTU model comprises a parallel elastic body (PE) and the elastic The body is involved when CE extends beyond the optimal length l _opt . In addition, the buffer elastic body (BE) ensures that the leg geometry / dimensions prevent the CE from collapsing when the MTU contracts to such an extent that the MTU will sag. Note that BE is just a numerical tool that allows MTU to represent relaxed muscles-for example, a relaxed GAS when the knee is bent too much. However, BE does not generate force outside the MTU.

Table 3 shows the individual MTU parameters. All parameters were estimated from Yamaguchi et al. (Yamaguchi, GT, Sawa, AG-U., Moran, DW, Fessler, MJ, Winters, JM, 1990. A survey of human musculotendon actuatorparameters. In: Winters, J., Woo, S. -Y. (Eds.), Multiple Muscle Systems: Biomechanics and Movement Organization. Springer- Verlag, New York, pp. 717-778). The maximum isometric force (also referred to as the maximum isometric (contraction) force) F _max is estimated from the individual or grouped muscle-physiological cross sections where a force of 25 N / cm ² is assumed. Maximum of contraction velocity v _max is set to 12l _opt / sec to 6l _opt / sec, against moderate fast-twitch for the slow-twitch. The optimal CE length l _opt and SE slack length (eg, the length at which slacking begins to transmit force) l _slack reflects the length of muscle fibers and tendons.

Details on how CE and SE were modeled can be found in the literature by Geyer et al. (Geyer, H., Seyfarth, A., Blickhan, R., 2003. Positive force feedback inbouncing gaits? Proc. R. Soc. Lond. B 270, 2173-2183). The CE force F _CE = AF _max f _l (l _CE ) f _v (v _CE ) is the muscle activity A, the CE force-length relationship f _l (l _CE ), and the CE force-velocity relationship f _v It is the product of (v _CE ). Based on this product approach, the MTU dynamics are calculated by integrating the _CE velocity v _CE obtained by inverting f _v (v _CE ). _{F SE = F CE + F PE} -F BE, f v (v CE) = a _{(F SE -F PE + F BE} ) / (AF max f l (l CE)). This equation has a numerical critical point during muscle extension (or muscle extension) as F _SE -F _PE approaches zero. In order to increase the speed of the simulation, this critical point introduces f _v (v _CE ) into the force generation F _PE of the parallel elastic body, ie F _PE ~ (l _CE −l _opt ) ² f _v (v _CE ). Note that PE is involved outside the normal behavior of the model and, like BE, plays a minor role in muscle dynamics during normal walking. However, this approach yields f _v (v _CE ) = (F _SE + F _BE ) / (AF _max f _l (l _CE ) + F _PE ), which integrates numerically using coarse time steps. It is possible. This approach is useful for speeding up the simulation of the model, but was also very useful when muscle dynamics were emulated on a PC board with constant and limited time resolution.

The MTU has common parameters and individual parameters. The common parameters are: time constant of excitement-contraction coupling: t _ecc = 0.01, width of CE force-length relationship: w = 0.56 l _opt and residual force coefficient: c = 0.05, CE Eccentric force enhancement of force-velocity relationship: N = 1.5 and shape factor of the relationship: K = 5, and SE reference strain: ε _ref = 0.04 (for details, see Geyer, H., Seyfarth, A., Blickhan, R., 2003. Positive force feedback in bouncing gaits? See Proc. R. Soc. Lond. B 270, 2173-2183). Common parameters include PE calibration strain: ε _PE = w (in this case F _PE = F _max (l _CE / l _opt −1) ² / ε _PE ² f _v (v _CE )), BE stationary length l _min = l _opt −w and its calibration compression (or reference compression): ε _BE = w / 2 (in this case, F _BE = F _max [(l _min −l _CE ) / l _opt ] ² / ε _PE ² ) There is also. Individual MTU attachment parameters can be easily obtained from the literature and identify each muscle or group of muscles. These values are listed in Table 4.

Musculoskeletal connections and mass distribution. The MTU connects to the bone (skeleton) by straddling them over one or two joints. For ankle and knee joints, the transformation from muscle force F _m to joint torque τ _m is modeled using a variable lever arm r _m (ψ) = r ₀ cos (ψ−ψ _max ). Here, [psi is joint angle, angle when [psi _max is _{the r m} has reached its maximum value, is _{τ m = r m (ψ)} F m. For the hip joint, simply r _m (ψ) = r ₀ is assumed. On the other hand, for ankle and knee joints, the MTU length change Δl _m is modeled as Δl _m = ρr [sin (ψ−ψ _max −sin (ψ _ref −ψ _max )], and for the hip joint, Δl _m is modeled as Δl _m = ρr (ψ−ψ _ref ), where the reference angle ψ _ref is the joint angle when l _m = l _opt + l _slack , and the coefficient ρ is the wing angle of the muscle (Pennation angle) factor (or expectation of winged angle), ensuring that the fiber length of the MTU stays within the physiological range throughout the range of operation of the joint. Specific parameters for muscles and joints are listed in Table 4. These values are experimental evidence (Muraoka, T., Kawakami, Y., Tachi, M., Fukunaga, T., 2001. Muscle fiber and tendon length changes in the human vastuslateralis during slow pedaling.J. Appl.Physiol. 91, 2035-2040; M Force-length characteristics of in vivo human skeletal muscle. ActaPhysiol. Scand. 172, 279-285; Maganaris, C, 2003. Force-length characteristics of the in vivo human gastrocnemius muscle. Clin. Anat. 16, 215-223; Oda, T., Kanehisa, H., Chino, K., Kurihara, T., Nagayoshi, T., Kato, E., Fukunaga, T., Kawakami, Y., 2005. Invivo lenth-force int. J. Sport and Health Sciences 3, 245-252), or obtained by rough anatomical estimation. relationships on muscle fiver and muscle tendon complex in the tibialis anterior muscle.

The seven segments of the human body model are simple rigid bodies whose parameters are listed in Table 5. These values are derived from other modeling studies such as Gunther and Ruder (Gunther, M., Ruder, H., 2003. Synthesis of two-dimensional human walking: a test of the λ-model. Biol. Cybern. 89 , 89-106). These segments are connected by a rotating joint. As in the human case, these joints have free-moving ranges (70 ° <ψ _a <130 °, ψ _k <175 °, and ψ _h <230 °), and outside that angular range, the machine The equation's soft limit is involved, but this is modeled in the same way as a ground collision point. Each segment of the model has a different mass m _s , length l _s , and their local mass center characteristic distance d _{G, S} , and joint position d _J (measured from the distal end or end). _{, S} and inertia Θ _s .

Ground contact and joint limitations. Each foot of the biped animal model has contact points on the toes (toes) and the heel. The contact point (CP) is pushed back by vertical reaction force (vertical reaction force) F _y = −F _ref f ₁ f _v when colliding (or contacting) the ground. Similar to the muscular force, this force includes force-length relationship f ₁ (Δy _CP ) = Δy _CP / Δy _ref and force-speed relationship f _v (dy _CP / dt) = 1−dy _CP / dt / v _max ( 9). This product approach to modeling vertical floor reaction forces is similar to existing approaches that describe vertical floor reaction forces as the sum of a spring term and a nonlinear spring-damper term (Scott, S., Winter , D., 1993. Biomechanical model of the human foot: kinematics and kinetics during the stance phase of walking.J. Biomech. 26 (9), 1091-1104; Gerritsen, K., van den Bogert, A., Nigg, B., 1995. Direct dynamicssimulation of the impact phase in heel-toe running.J. Biomech. 28 (6), 661-668; Gunther, M., Ruder, H., 2003. Synthesis of two-dimensional humanwalking: a test of the λ-model. Biol. Cybern. 89, 89-106). However, by separating the spring and damper terms, the parameters of the contact model are reduced to two basic material properties: ground stiffness k = F _ref / Δy _ref and maximum relaxation speed v _max (the speed is And how fast the ground can recover to its original shape after deformation). For example, v _max = ∞ means that the collision (or contact) with the ground is completely elastic, in which case the ground always acts to push back the CP. Also, v _max = 0 means that the collision (or contact) with the ground is completely inelastic, in which case the ground pushes the CP back for a downward velocity, just like sand. , Do not push back for upward speed. The same collision model will be used to describe the mechanical soft limits (see previous section) of the joints of the model, with a soft limit stiffness of 0.3 Nm / deg and a maximum relaxation rate of 1 deg / sec. Please note that.

FIG. 9 illustrates a contact model according to one aspect of the present invention. In Figure 9, the contact when the contact point 920 is lower than _{y 0} has occurred (910). The vertical floor reaction force F _y is modeled as a product of a force-length relationship (f ₁ ) and a force-velocity relationship (f _v ), as in the case of muscle force. In this case, Δy _ref is dy / dt = 0. Is the degree of compression (or amount of compression) of the ground when F _y = F _ref, and dy _ref / dt is the maximum relaxation rate of the ground (small drawing). First, the horizontal floor reaction force (the horizontal component of the floor reaction force) F _x is modeled as sliding friction (having a sliding friction coefficient μ _sl ) proportional to F _y . However, the horizontal model switches to stiction (or static friction) 930 when the contact point 920 (velocity) decelerates below the minimum velocity v _lim . Among stiction 930, F _x also force - length relationship and force - but is modeled as the product of velocity relation, which is the interaction with the ground near the stiction reference point x ₀ can be in both directions To be a little different from the previous one. The model switches to sliding friction when F _x exceeds the stiction limit force μ _st F _y . Parameters F _ref = 815 N, Δy _ref = 0.01 m, dy _ref /dt=0.03 m / sec, Δx _ref = 0.1 m, dx _ref /dt=0.03 m / sec, v _lim = 0.01 m / sec , Μ _sl = 0.8, μ _st = 0.9.

During ground contact, horizontal reaction force is applied to CP in addition to vertical floor reaction force. Initially, this force is modeled as a dynamic friction force that counteracts the movement of the CP on the ground with the force F _x = μ _sl F _y . As the CP decelerates below the velocity v _lim , the horizontal reaction force is modeled as a stiction force calculated in a similar manner to calculating the vertical impact force (FIG. 9). Stiction changes to dynamic friction when the stiction force exceeds the limit force F _lim = μ _st F _y . Thus, depending on the transition conditions, both types of horizontal reaction force alternate with each other until the CP leaves the ground.

  This result suggests that dynamics and motor control cannot be considered separately in human walking. The neuromuscular model of human walking according to one aspect of the present invention is that the walking state is configured autonomously after the first push, and the ground (height) changes rapidly without any intervention. Endure and adapt to staircase walking. Central to this model's tolerance and adaptability is the model's reliability to muscle reflexes, which incorporate perceptual information (sensor information) on gait dynamics into leg muscle activity. The model reveals that, in the absence of CPG, in principle, the central input is not required to generate gait, and the nervous system and its mechanical environment. This suggests that a reflex input that continuously mediates the (physical environment) may override the central input in controlling normal human walking.

  Furthermore, the model results suggest that their continuous reflective inputs encode (encode) the principles of gait mechanics (or gait mechanisms). Current experimental and modeling studies on the role of spinal reflex during walking focus on the timing of swing and stance and their contribution to the generation of muscle strength in load-bearing extensors (Pang , MY, Yang, JF, 2000.The initiation of the swing phase inhuman infant stepping: importance of hip position and leg loading.J Physiol528 Pt 2, 389-404; Dietz, V., 2002.Proprioception and locomotor disorders. NatRev Neurosci 3 (10), 781-790; Ivashko, DG, Prilutski, B. L, Markin, SN, Chapin, JK, Rybak, IA, 2003.Modeling the spinal cord neural circuitry control cat cat hindlimb movement during locomotion. 629; Yakovenko, S., Gritsenko, V., Prochazka, A., 2004. Contribution ofstretch reflexes to locomotor control: a modeling study.Biol Cybern 90 (2), 146-155; Ekeberg, O., Pearson, K. , 2005. Computer simulation of stepping in thehind legs of the cat: an examination of mechanisms regulating J Neurophysiol 94 (6), 4256-4268; Maufroy, C, Kimura, H., Takase, K., 2008. Towards a general neural controller forquadrupedal locomotion. Neural Netw 21 (4), 667 -681; Donelan, JM, Pearson, KG, 2004. Contribution of sensory feedback to ongoing ankle extensoractivity during the stance phase of walking.Can J Physiol Pharmacol 82 (8-9), 589-598; Frigon, A., Rossignol, S., 2006. Experiments and models of sensorimotor interactions during locomotion. Biol Cybern 95 (6), 607-627; Gray, MJ, Nielsen, JB, Mazzaro, N., Sinkjaer, T., 2007. Positive force feedbackin human walking. J Physiol 581 (1), 99-105). The contribution of reflection to load bearing began with coupling positive force feedback to the dynamics underlying the walking system (Prochazka, A., Gillard, D., Bennett, D., 1997. Positive forcefeedback control of muscles. J. of Neurophys. 77, 3226-3236; Geyer, H., Seyfarth, A., Blickhan, R., 2003. Positive force feedback in bouncing gaits? Proc. R. Soc. Lond. B 270, 2173 -2183). There seems to be no systematic investigation of the idea of coding the principles of walking mechanics in a motor control system. Some of the muscle reflexes (nerves) implemented in the human body model were simple means for transitioning to periodic movements (trunk balance, swing leg initiation), but compliant stance leg behavior ( Blickhan, R., 1989.The spring-mass model for running and hopping.J. Of Biomech. 22, 1217-1227; McMahon, T., Cheng, G., 1990.The mechanism ofrunning: how does stiffness couple with speed J. of Biomech. 23, 65-78; Geyer, H., Seyfarth, A., Blickhan, R., 2006. Compliant leg behavior explains thebasic dynamics of walking and running. Proc. R. Soc. Lond. B 273 , 2861-2867), segmented chains (Seyfarth, A., Gunther, M., Blickhan, R., 2001. Stable operation of an elastic three-segmented leg. Biol. Cybern. 84, 365-382; Gunther, M., Keppler, V., Seyfarth, A., Blickhan, R., 2004. Human leg design: optimal axialalignment under constraints. J. Math. Biol. 48, 623-646), and pull back of the free leg (Herr) , H., McMahon, T., 2000. A trotting horse m odel. Int. J. Robotics Res. 19, 566-581; Herr, H., McMahon, T., 2001. A galloping horsemodel. Int. J. Robotics Res. 20, 26-37; Herr, HM, Huang, GT, McMahon, TA, Apr 2002.A model of scale effects in mammalian quadrupedal running.J ExpBiol 205 (Pt 7), 959-967; Seyfarth, A., Geyer, H., 2002.Natural control ofspring-like running- optimized self-stabilization.In: Proceedings of the 5thinternational conference on climbing and walking robots.Professional Engineering Publishing Limited, pp. 81- 85; Seyfarth, A., Geyer, H., Herr, HM, 2003.Swing-leg retraction: a It was mainly the reflection of the stance phase that encoded the above-mentioned gait dynamics and control principles including J. Exp. Biol. 206, 2547-2555). Based on these functional reflexes, the model not only converges to known joint angle and torque trajectories of human walking, but also predicts several individual muscle activity patterns observed in walking experiments. The agreement between predicted and observed muscle activity is that the principle of gait mechanics can play a greater role in motor control than previously anticipated, in which case Reflection suggests combining these principles with neural networks involved in walking.

  In a preferred embodiment, the neuromechanical model of the present invention is implemented as a muscle reflex controller for a powered short leg brace. This embodiment is an adaptive muscle reflex controller that utilizes an ankle plantar flexor muscle with hill-type muscles having positive force feedback reflex based on simulation studies. The parameters of this model match the torque-angle profile of the human ankle obtained from measurements when an uninjured subject with a balance of weight and height walks on the level ground at 1 m / sec. So adapted. Using this single parameter set, clinical trials were conducted on leg amputees walking on the horizontal ground, uphill and downhill. During these tests, the ankle function of the brace in response to ground gradient changes in a manner equivalent to an uninjured subject without having to deal with the problem of making a clear terrain determination. Was observed to adapt. Specifically, the energy supplied by the brace has a direct correlation with the ground inclination angle. This study highlights the importance of neuromuscular controllers to enhance the adaptability of powered prostheses (prosthetics) that move over changing terrain surfaces.

  In order to create an adaptive controller, a neuromuscular model with a positive force feedback reflex configuration as the basis for control of the present invention was used as part of a control system for a powered short leg brace. The controller presented here determines the physical torque for commanding the ankle joint (ankle joint) using the ankle-foot complex model. In this model, two virtual actuators are provided at the ankle joint. In the case of plantar flexion torque, the actuator is a hill-type muscle with a positive force feedback reflex configuration. This configuration models the reflex muscle response resulting from the combination of afferent signals from the muscle spindle and Golgi tendon spindle. In the case of dorsiflexion torque, the impedance is provided by a virtual rotary spring-damper.

  The parameters of this neuromuscular model are input the measured ankle torque of an undamaged subject walking at a target speed of 1.0 m / sec and the measured movement of the undamaged subject. Was adapted (or set) by an optimization procedure to best match the output torque of the model as given. Using this neuromuscular model-based prosthetic controller, a torque command was provided to a powered short leg brace worn by a limb amputee. This control scheme was evaluated using two criteria. First, the indirect foot torque and ankle joint angle profile of the prosthesis are determined by comparing the foot indirect torque of the undamaged subject (comparable to the prosthesis) walking on the horizontal ground at the target walking speed and The controller was tested for the ability to produce a qualitative match with the ankle angle profile. The second performance criterion is that of a controller that exhibits a biologically consistent trend of increasing the net work of the walking cycle when the slope of the walking slope is increased without changing the controller parameters. It was ability. Since it is difficult to detect changes in ground inclination using common sensors, a controller with the inherent ability to adapt to those changes is valuable.

FIGS. 10A-10C are diagrams of an exemplary embodiment physical system (FIG. 10A), drive train (drive system) of an ankle foot orthosis used in the preferred embodiment (FIG. 10B). And a dynamic model (C in FIG. 10). The short leg orthosis used in this study was developed by iWalk, LLC. This bracelet is a successor to a series of prototypes developed at the MIT Media Lab's Biomechatronics Group, which are described in US patent application Ser. No. 12 / 157,727 filed Jun. 12, 2008. The entire contents of which application are hereby incorporated by reference. The brace is a fully independent device and has the weight (1.8 kg) and size of an intact biological ankle-foot complex. FIG. 10A shows a housing (housing) 1005 for housing a motor, a power transmission device (transmission), and an electronic circuit, an ankle joint (foot joint) 1010, a foot 1015, a unidirectional longitudinal leaf spring (parallel). A leaf spring (or parallel leaf spring) 1020 and a series leaf spring 1025 are shown. 10B includes a timing belt 1030, a pin joint main housing 1035, a motor 1040, a ball screw 1045, an ankle joint (foot joint) 1010, a ball nut 1050, a pin joint (series spring) 1055, and a foot motion indicator 1060. It is shown. 10 includes a parent link 1065, a motor 1040, a power transmission device (transmission) 1070, a series spring 1025, a unidirectional longitudinal spring (or a unidirectional parallel spring). ) 1020, foot 1015, series spring moving arm r _s 1075, spring rest length 1080, and SEA 1085 are shown. The rotating elements in the physical system are shown as linear equivalents for clarity in the model diagram.

  The ankle joint is a rolling bearing structure that couples the lower foot structure to the upper leg structure, and includes a pyramid-shaped prosthetic attachment on the leg structure for attaching to the socket of a person who has lost the foot. Yes. The foot is, for example, a passive thin Flex-Foot ™ (Osur ™) to minimize the impact on the ground contact experienced by a person who has lost the foot. Unidirectional leaf springs or longitudinal springs operate across the ankle joint and engage when the ankle and foot are in a vertical relationship. The spring operates in parallel (or simultaneously) with a powered (or motorized) drive train to provide the passive function of the Achilles tendon. The motorized drivetrain is a motorized (or motorized) interlocking device that traverses the ankle joint, as shown in FIG. 10B. This consists of a brushless motor (Powermax EC-30, 200 watts, 48V, Maxon) operating at 24V from the end of the upper leg, a belt drive with a reduction ratio of 40/15, and a linear ball screw with a pitch of 3 mm ( linear ball screw). At this operating voltage, the theoretical maximum torque that can be generated by the motor via the drive train is about 340 Nm.

The foot, the series spring, i.e., Kevlar-composite (kevlar composite) leaf spring, connected to the directional dependence of the ball nut having a moment arm r _s the foot. Therefore, the effective rotary stiffness of the series spring, evaluated by locking the drivetrain and applying torque around the ankle, is 533 N · m / radian for positive torque. Yes, for negative torque is 1200 N · m / radian, where positive torque (ie plantar flexion torque) tends to compress the series spring as shown in FIG. . Drivetrain and series spring are series-elastic actuators (SEA. GAPratt and MM Williamson, "Series elastic actuators," Proceedingson IEEE / RSJ International Conference on Intelligent Robots and Systems, Pittsburgh, pp. 399-406, 1995). An outline of the arrangement of these components is shown in FIG.

  sensor. Hall effect angle sensors at the ankle joint are the primary control input and have a range of -0.19 to 0.19 radians, where zero is the foot is perpendicular to the leg (or shin) Corresponding to that. The joint angle (joint angle) is measured by a linear Hall effect sensor (Allegro A1395) mounted on the main housing. The sensor is placed near a magnet that is firmly connected to the foot structure so that the magnetic axis is in contact with the arc that is the path of movement of the magnet. Thanks to this arrangement, the strength of the magnetic field at the sensor position changes as the magnet rotates across the sensor. The strain gauge is placed inside the pyramid-shaped prosthesis fixture to allow measurement of torque at the ankle joint. Strain gauges arranged in series springs enable detection of the output torque of the powered drive train, thereby enabling SEA closed loop force control. The motor itself has a Hall effect commutation sensor and is fitted with an optical shaft encoder that enables advanced brushless motor control technology.

  Microcontroller. Overall control and communication of the ankle brace is provided by a single chip 16 bit DSP oriented microcontroller, dsPIC33FJ128MC706 from Microchip Technology Incorporated. The microcontroller executes 40 million instructions per second and has 128 kilobytes of flash program memory and 16384 bytes of RAM. The microcontroller provides sufficient computing power to support real-time control.

  Motor control. Another 16-bit dsPIC33FJ128MC706 was used as a dedicated motor controller. The high computational load and high speed requirements of modern brushless motor control systems have motivated the adoption of this architecture, along with task separation from the real-time requirements of the main microcontroller. The motor command was provided virtually instantaneously by the high speed digital link between the main microcontroller and the motor microcontroller.

  Wireless interface. For development and data collection, the microcontroller's high-speed serial port is assigned to external communication. This port can be used directly via a cable, or various wireless communication devices can be attached to this port. In this study, sensor information and internal state information at 500 Hz are transmitted (remotely) via a serial port at 460 kbaud and transmitted via an IEEE 802.11g wireless local area network device (Lantronix Wiport).

  Battery (battery). All power for the prosthesis was provided by a 0.22 kg lithium polymer battery with an energy density of 165 watt hours / kg. This battery was able to provide a daily power requirement including a 5000 step drive walk (powered walk).

  Selection of optimal mechanical parts (machine parts). Meeting the requirements of mass, size (size), torque, speed, energy efficiency, impact resistance, and operation with little sound is an important issue. Of particular importance is the modeling and optimization of the drive train to generate biological torque and walking motion. Several effects of motor selection, overall speed transmission ratio (or transmission ratio), series elastic springs, and longitudinal springs are “On the Design of a Powered Ankle-Foot Prosthesis: The Importance of Series and Parallel Elasticity, “IEEE Robotics & Automation Magazine, pp. 52-59, September 2008”.

  Control architecture. The purpose of the control architecture is to command an ankle torque that is suitable for the walking cycle of the lost leg determined from available sensor measurements of the prosthetic ankle state. The controller uses the neuromuscular model of the human ankle-foot complex to determine the appropriate torque. In this model, the hinge joint (or hinge joint) representing the human ankle joint is actuated by two competing virtual actuators. These actuators act as either a unidirectional plantar flexor, a hill-type muscle model, or a bi-directional proportional differential position controller and a unidirectional virtual rotary spring-damper, depending on the walking stage. It is a flexor. The finite state machine maintains an estimate of the walking stage of the foot lost person. Depending on this estimated walking stage, either or both of the virtual actuators will generate torque at the virtual ankle joint. In this case, the net virtual torque is used as an ankle torque command to the hardware of the prosthesis. The physical torque at the ankle joint is generated by both a powered drive train and a longitudinal spring. An ankle joint angle sensor is used to determine the torque generated by the longitudinal spring and the remaining desired torque is commanded by the motor controller.

  Top level state machine control. The top level control of the prosthesis is performed by a finite state machine synchronized to the walking cycle. During walking, two states are recognized: the swing phase and the stance phase. Prosthetic sensor inputs (ankle torque, ankle angle, and motor speed estimated from pyramid-shaped strain gauges) are continuously observed to determine state transitions. These state transition conditions were determined experimentally. FIG. 11 shows the state machine operation and transition conditions. The virtual actuator that acts as the dorsiflexor and plantar flexors generates torque depending on the walking state estimate from the state machine.

In FIG. 11, the free leg state 1110 is visually shown as SW 1120, the stance state 1130 is controlled plantar flexion (CP) 1140, controlled dorsiflexion (CD) 1150, and powered sole It is divided into bending (PP) 1160. The state transitions 1170 and 1180 are determined using the prosthetic ankle joint torque T _P measured by a pyramidal strain gauge and the prosthetic ankle joint angle θ.

  The transition to the swing phase in which the foot leaves the ground is to reduce the total ankle joint torque measured using a pyramidal strain gauge to less than 5 N · m or prevent saturation of the angle sensor. The detected ankle joint angle θ is detected by decreasing to less than −0.19 radians (or less than −0.19 radians). Positive torque is defined as the actuator torque that acts to bend the ankle joint (ankle), and a positive angle corresponds to dorsiflexion. To prevent premature state transitions, the ankle torque generated during the stance phase must exceed 20 N · m for those transitions to be possible. Furthermore, a buffer time of 200 ms (milliseconds) provides a minimal time frame during the stance phase. The transition to the stance phase at the time of heel contact is detected when the torque measured using a pyramidal strain gauge decreases to less than -7 N · m (or less than -7 N · m).

  A block diagram of an exemplary embodiment of a control system for ankle foot orthosis according to this aspect of the invention is shown in FIG. In FIG. 12, a neuromuscular model 1210, a longitudinal spring (or parallel spring) model 1220, a lead compensator 1230, a friction compensator 1240, a motor controller 1250, and a mechanical model (shown as C in FIG. 10). A prosthesis 1260 is shown.

を用いて、神経筋モデルからのトルク指令τ_ｄを生成する。この所望の足関節トルクは、現在のトルク指令を得るためにトルク制御システムを通じて人工装具のアクチュエーターに送られる。このトルク制御システムの３つの主要な構成要素は、フィードフォワードゲインＫ_ｆｆ、進み補償器、及び、摩擦補償項である。所望のアクチュエータートルクτ_ｄ，SEAを得るために、人工装具の足関節トルクτ_ｐに対する縦向きばねの寄与が所望の足関節トルクから差し引かれる。次に、閉ループトルクコントローラが、測定されたアクチュエータートルクτ_SEAを用いて所望のアクチュエータートルクを生成するように制御する。最後に、摩擦補償項によって追加のトルク値τ_ｆが生成され、これが、閉ループトルクコントローラの出力に追加される。 Measured joint condition of the prosthesis

Is used to generate a torque command τ _d from the neuromuscular model. This desired ankle torque is sent through the torque control system to the prosthetic actuator to obtain the current torque command. The three main components of this torque control system are a feed forward gain K _ff , a lead compensator, and a friction compensation term. In order to obtain the desired actuator torque τ _{d, SEA} , the contribution of the longitudinal spring to the prosthetic foot joint torque τ _p is subtracted from the desired foot joint torque. Next, a closed loop torque controller controls to generate the desired actuator torque using the measured actuator torque τ _SEA . Finally, an additional torque value τ _f is generated by the friction compensation term and added to the output of the closed loop torque controller.

  FIGS. 13A-C are plots of prosthetic torque over a complete walking cycle (from heel contact to heel contact of the same foot) for the three walking conditions. These are the ground (A in FIG. 13), the upslope (B in FIG. 13), and the downslope (C in FIG. 13). For each of these, the mean 1305, 1310, 1315 (thin line) ± standard deviation (dashed line) of the command torque estimated using the measured SEA torque contribution and the torque of the prosthesis, and the longitudinal spring Angle-based estimates 1320, 1325, 1330 (thick lines) of torque contribution are shown. Vertical lines (broken lines) 1335, 1340, and 1345 indicate the end of the stance phase.

  Dorsal flexor model. FIG. 14 illustrates an exemplary embodiment of a musculoskeletal model implemented in a prosthetic microcontroller, including a spring-damper connection mechanism (A in FIG. 14) to a hill-type muscle model and a two-joint ankle model. Includes a detailed hill-type muscle model (B in FIG. 14) and the geometry / arrangement of the skeletal connection mechanism of the muscle model (C in FIG. 14), which includes variable moment-arms and muscles An angular coordinate system for the model is included. 14A and 14C show mechanical representations of a dorsiflexor (spring-damper) 1405, a plantar flexor muscle (MTC) 1410, a foot 1415, a leg (shin) 1420, and a heel 1425. FIG.

であって、

という関係を有する仮想回転ばね−ダンパーとして実施される。 The dorsiflexor muscle of A of FIG. 14 is a dorsiflexor actuator. This represents the Tibialis Anterior and other biological dorsiflexors. This dorsiflexor has a set point (set value)

Because

This is implemented as a virtual rotary spring-damper having the relationship

は足関節角速度である。立脚期の間、Ｋ_Ｐの値は、水平な地面上の正常な歩行時の生物学的足関節の立脚期の挙動に最良に合致するように、他の筋肉モデルパラメータと共に最適化された。減衰係数Ｋ_Ｖは、立脚期の間は、前足が足底接地時に地面からはね返るのを防ぐために、実験に基づいて５Ｎｍ-s／ラジアンに調整された。また、立脚期の間、背屈筋は、生物学的筋肉の一方向特性をまねるべく背屈トルクを提供するためだけに作用する。さらに、足が脚（またはすね）に垂直になった結果、立脚期中に背屈筋によって生成されたトルクがゼロに下がると、背屈筋の動作が、立脚期の残りの期間停止させられる。したがって、背屈筋は、立脚期の初期においてトルク生成に寄与するだけであり、この初期の期間では、人間の背屈筋が重要な役割を果たすことが知られている（J. Perry, Gait Analysis: Normal and Pathological Function, New Jersey: SLACK Inc.,1992, Chapter 4, pp. 55-57）。遊脚期には、背屈筋は、位置コントローラとして機能して、足を設定ポイント

まで動かす。この場合、Ｋ_Ｐ＝２２０Ｎ・ｍ／ラジアンの利得、及びＫ_Ｖ＝７Ｎ・ｍ・ｓ／ラジアンの減衰係数によって、遊脚期の初期にすぐに足を地面から離すことが可能になる。 Where K _P is the spring constant, K _V is the damping constant, θ is the ankle joint angle, and

Is an ankle joint angular velocity. During the stance phase, the value of K _P was optimized along with other muscle model parameters to best match the stance phase behavior of the biological ankle during normal walking on the horizontal ground. Damping coefficient K _V during the stance phase, to prevent the forefoot that bounce off the ground when the sole ground and adjusted to 5 Nm-s / radian empirically. Also during the stance phase, the dorsiflexor acts only to provide dorsiflexion torque to mimic the unidirectional characteristics of biological muscle. In addition, when the foot is perpendicular to the leg (or shin) and the torque generated by the dorsiflexor during the stance phase falls to zero, the dorsiflexor movement is stopped for the remainder of the stance phase. Thus, the dorsiflexor only contributes to torque generation in the early stance phase, and it is known that the human dorsiflexor plays an important role in this early period (J. Perry, Gait Analysis: Normal and Pathological Function, New Jersey: SLACK Inc., 1992, Chapter 4, pp. 55-57). During the swing phase, the dorsiflexor acts as a position controller and sets the foot point

Move up. In this case, the gain of K _P = 220 N · m / radian and the attenuation coefficient of K _V = 7 N · m · s / radian allow the foot to be released immediately from the ground at the beginning of the swing phase.

  Plantar flexor model. The virtual plantar flexor muscles of FIGS. 14A to 14C include a muscle-tendon complex (MTC) representing a human plantar flexor muscle combination. This MTC is the “Biomechanical design of a powered ankle-foot prosthesis” by SK Au, J. Weber and H. Herr (Proc. IEEE Int. Conf. On Rehabilitation Robotics, Noordwijk, The Netherlands, pp. 298-303, June 2007. MTC is described in detail in this document. The MTC is composed of a contractile element (CE) that models muscle fibers and a serial element (SE) that models tendons. The contractile element has three unidirectional elements: a hill-type muscle with a positive force feedback reflex mechanism, a parallel (or parallel) elastic element with a high elastic limit, and a low elastic limit (or buffer) ) Composed of parallel (or parallel) elastic elements. A series element that is a non-linear unidirectional spring representing the Achilles tendon is connected in series with the contractile element. The geometry / arrangement of the attachment mechanism for attaching the muscle-tendon complex to the ankle model is non-linear, complicating the calculation of torque resulting from the actuator force.

と定義する。ｌ_SEは直列要素の長さ、ｌ_slackはその静止長（レスト長）であり、該直列要素は、「H. Geyer, A. Seyfarth, R. Blickhan, "Positive force feedback inbouncing gaits?," Proc. R Society. Lond. B 270, pp. 2173-2183, 2003」に記載されている非線形ばねとして規定される。すなわち、

である。ここで、F_maxは、筋肉が及ぼすことができる最大アイソメトリックフォース（最大等尺性（収縮）力）である。「H. Geyer, A. Seyfarth, R. Blickhan, "Positive force feedback inbouncing gaits?," Proc. R Society. Lond. B 270, pp. 2173-2183, 2003」によれば、この二次形式は、一般的にモデル化されている区分的に指数−線形の腱剛性曲線の近似として使用された。この近似は、モデルパラメータの数を少なくするためになされた。 A series elastic element of plantar flexor muscles. The series elastic element (SE) is described in “H. Geyer, A. Seyfarth, R. Blickhan,“ Positive force feedback inbouncing gaits ?, ”Proc. R Society. Lond. B 270, pp. 2173-2183, 2003”. Acts as a tendon connected in series to the muscle contractile element as is. Let ε be the tendon strain,

It is defined as l _SE is the length of the series element, l _slack is the rest length (rest length), and the series element is “H. Geyer, A. Seyfarth, R. Blickhan,“ Positive force feedback inbouncing gaits ?, ”Proc R Society. Lond. B 270, pp. 2173-2183, 2003 ”. That is,

It is. Here, F _max is the maximum isometric force (maximum isometric (contraction) force) that the muscle can exert. According to "H. Geyer, A. Seyfarth, R. Blickhan," Positive force feedback inbouncing gaits ?, "Proc. R Society. Lond. B 270, pp. 2173-2183, 2003, this secondary form is Used as an approximation to the piecewise exponential-linear tendon stiffness curve that is commonly modeled. This approximation was made to reduce the number of model parameters.

によって与えられる。ヒルタイプの筋肉の力−長さ関係f_L（l_CE）は、鐘形（ベル形）の曲線であり、

によって与えられる。ここで、l_optは収縮性要素の長さl_CEであり、この場合に、筋肉は、最大のアイソメトリックフォース（最大の等尺性（収縮）力）F_maxを提供することができる。パラメータｗは、鐘形曲線の幅であり、パラメータｃは、該鐘形の極値の近くの該曲線の値を表しており、

である。CEの力−速度関係f_V（V_CE）は、次のヒル方程式

で表される。ここで、v_max（＜０）は筋肉の最大収縮速度であり、v_CEは繊維収縮速度であり、Ｋは曲率定数（curvature constant）であり、Ｎは、

となるように、（F_maxで正規化された）無次元筋力（dimensionless muscle force）を規定する。 Contractor element of plantar flexor: The contractor element (CE) of the virtual actuator of plantar flexor muscle of FIG. 14B is a hill type muscle model with a positive force feedback reflex mechanism. This can be done as described in “H.Geyer, H. Herr,“ A muscle-reflex model that encodes principles of leggedmechanics predicts human walking dynamics and muscle activities ”(submitted for publication). Includes active muscle fibers to produce and two parallel elastic elements. Hill type muscle fibers generate unidirectional forces. This force is a function of muscle fiber length l _CE , velocity V _CE , and muscle activation (or muscle activity) A. The resulting force F _MF is described in “H. Geyer, A. Seyfarth, R. Blickhan,“ Positive force feedback inbouncing gaits ?, ”Proc. R Society. Lond. B 270, pp. 2173-2183, 2003”. As

Given by. The hill-type muscle force-length relationship f _L (l _CE ) is a bell-shaped curve,

Given by. Here, l _opt is the length l _CE of the contractile element, in which case the muscle can provide the maximum isometric force (maximum isometric (contraction) force) F _max . The parameter w is the width of the bell-shaped curve and the parameter c represents the value of the curve near the extreme value of the bell-shaped curve,

It is. The CE force-velocity relationship f _V (V _CE ) is

It is represented by Where v _max (<0) is the maximum muscle contraction rate, v _CE is the fiber contraction rate, K is the curvature constant, and N is

Define dimensionless muscle force (normalized by F _max ) so that

によって与えられる。弾性限界が小さいバッファー並列弾性要素（LPE）も、（出版のために提出された）「H.Geyer, H. Herr, "A muscle-reflex model that encodes principles of leggedmechanics predicts human walking dynamics and muscle activities」に基づいて含まれている。これは、次の非線形ばねの形態で与えられた。

したがって、全体の足底屈筋力は、

によって記述される。ここで、F_CEは収縮性要素によって生成される力である。CEとSEは直列であるので、F_CE＝F_SE＝F_MTCという式が成立する。 According to “H.Geyer, H. Herr,“ A muscle-reflex model that encodes principles of leggedmechanics predicts human walking dynamics and muscle activities ”(submitted for publication), the elastic limit placed in parallel with CE The force-length relationship of a large parallel elastic element (HPE) is

Given by. A buffer parallel elastic element (LPE) with low elastic limit is also “H. Geyer, H. Herr,“ A muscle-reflex model that encodes principles of legged mechanicals predicts human walking dynamics and muscle activities ” Included on the basis of. This was given in the form of a non-linear spring:

Therefore, the total plantar flexor strength is

Described by. Where F _CE is the force generated by the contractile element. Since CE and SE are in series, the equation F _CE = F _SE = F _MTC is established.

  Reflection mechanism. The activation (activity) A of the contractile element is "H. Geyer, H. Herr," Amuscle-reflex model that encodes principles of legged mechanics predicts humanwalking dynamics and muscle activities, " And “H. Geyer, A. Seyfarth, R. Blickhan,“ Positive force feedback in bouncing gaits ?, ”Proc. R Society. Lond. B 270, pp. 2173-2183, 2003”. , Generated using the positive force feedback reflection configuration shown in FIG. FIG. 15 illustrates an exemplary embodiment of a virtual plantar flexor reflex configuration, which includes relationships between plantar flexor components of ankle joint angle, muscle strength, and ankle torque.

As shown in FIG. 15, the feedback loop includes a stance phase switch for preventing plantar flexor strength from being generated during the swing phase. During the stance phase, the plantar flexor strength F _MTC is multiplied by the reflection gain (reflection gain) Gain _RF , delayed by Delay _RF , and then the offset stimulation PRESTIM is added to obtain the nerve stimulation signal. The stimulus is limited to a range of 0 to 1 and is passed through a low-pass filter having a time constant T to stimulate muscle excitation-contraction coupling. The resulting signal is used as the degree of activation (activity) in equation (4) with the initial value PreA. Furthermore, according to “H. Geyer, H. Herr,“ A muscle-reflex model that encodes principles of legged mechanics predicts human walking dynamics and muscle activities ”(submitted for publication), the suppression gain Gain _SUPP Implemented to help prevent two actuators from competing with each other. Here, the torque generated by the dorsiflexor is reduced by Gain _SUPP · F _MTC or until its value drops to zero.

ここで、ρは、筋繊維の羽状角（pennation angle）を表すスケーリングファクタであり、φ_refは、負荷がない状態におけるl_CE＝l_optのときの足関節角度である。 Plantar flexor geometry / arrangement and implementation. In the framework of the muscle model, the ankle joint angle θ _foot is defined as shown in FIG. By using this angle as an input to the model, the length of the muscle-tendon complex can be derived from the "H. Geyer, H. Herr," A muscle-reflex model that encodesprinciples of It is calculated by the following formula as described in "legged mechanics predicts human walking dynamics and muscleactivities".

Here, ρ is a scaling factor representing the wing angle (pennation angle) of muscle fibers, and φ _ref is the ankle joint angle when l _CE = l _{opt in} a state where there is no load.

The fiber length l _CE can be calculated using l _CE = l _MTC -l _SE , where l _SE is given the current value of F _CE = F _SE = F _MTC from muscle dynamics It is obtained from the inverse function of (3). The fiber shrinkage rate V _CE can then be obtained using differentiation. This produces a first order differential equation determined by the dynamics of the neuromuscular model. Given a time history of θ _foot and initial conditions, this equation can be solved for _FMTC . However, since integrals are computationally more robust than derivatives, “H. Geyer, H. Herr,“ A muscle-reflex model that encodes principles of legged mechanics predicts human walking dynamics and muscle activities ” The integral form of this implementation was used to solve for F _MTC as described in.

によって与えられる。ここで、R(θ_foot)は、筋肉取付具から足関節モデルまでに生じる可変モーメントアームである。この関係は図６のグラフに示されている。それゆえ、結局のところ、足底屈筋モデルを、単一入力θ_footを単一出力T_plantarに結合する動的システムとして取り扱うことができる。 Relationship between F _MTC and ( _consecutive ) plantar flexor contribution T _plantar to the ankle torque when the radius of the fitting is r _foot and the angle at which the maximum muscle-tendon moment arm is achieved is φ _max Is

Given by. Here, R (θ _foot ) is a variable moment arm generated from the muscle fitting to the ankle joint model. This relationship is shown in the graph of FIG. Therefore, after all, the plantar flexor model can be treated as a dynamic system that couples a single input θ _foot to a single output T _plantar .

  Determination of neuromuscular model parameters. The plantar flexor model expresses all of the biological plantar flexors collectively (that is, together). Similarly, dorsiflexors represent all biological dorsiflexors. In this study, joint and torque measurements were taken only at the ankle joint. As a result, the state of articulated muscles such as the gastrocnemius could not be accurately predicted. Therefore, the plantar flexor muscle was based on the soleus muscle, which is the main single joint plantar flexor muscle of humans. Therefore, most of the plantar flexor muscle parameter values for the soleus muscle are “H. Geyer, H. Herr,“ A muscle-reflex model that encodes principles of legged mechanics predicts human walking dynamics and muscleactivities, ”. Are submitted for the purpose). However, not only the parameters of the dorsiflexor but also some parameters of the plantar flexor were predicted to be greatly affected by this collective model or were not well known in biology. These six parameters are adapted using a combination of genetic algorithms and gradient descent to best fit the neuromuscular model to the gait data of an uninjured subject. (Or set).

Table 6 shows parameter values that are not optimized.

  Collection of subject data without limb loss. Kinetic and kinematic gait data were collected from a Gait Laboratory of Spaulding Rehabilitation Hospital, Harvard Medical School, in a study approved by the Spalding committee on the Use of Humans as Experimental Subjects. (H. Herr, M. Popovic, “Angular momentum in human walking,” The Journal of Experimental Biology, Vol. 211, pp 487-481, 2008). A healthy adult male (81.9 kg) was required to walk at a slow walking speed in a 10 m passage at the Motion Capture Laboratory after informed consent was obtained.

  Motion capture was performed using a VICON 512 motion capture system with 8 infrared cameras. Reflective markers were placed at 33 locations on the subject's body so that the infrared camera could track their position during the experiment. The cameras were able to track a given marker with an accuracy within about 1 mm, operating at 120 Hz. These markers were placed on the following body landmarks to track the lower body: bilateral anterior superior iliac spines, posterior superioriliac spines, lateral femur Lateral femoral condyles, lateral malleoli, forefoot and heel. A rod (or wand) was placed over the tibia and femur and a marker was attached to the rod on the mid-femur mid-axis of the tibia and femur. Markers were also placed on the upper sternum, clavicle, C7 and T10 vertebrae, head, both shoulders, elbow, wrist joint (wrist joint):

  The floor reaction force was measured using two staggered force plates (model number 2222 or OR6-5-1 from Advanced Mechanical Technology Inc. (Watertown, MA, USA)) incorporated in the passage. The accuracy of these force plates for measuring the floor reaction force and the center of pressure is about 0.1 N and 2 mm, respectively. Force plate data was collected at 1080 Hz and synchronized with VICON motion capture data. Joint torque was calculated from floor reaction force and joint kinematics using a modified version of the standard inverse dynamics model. Oxford Metrics (UK) VICON Bodybuilder (3D biological model generation tool) was used to perform inverse dynamics calculations.

  Six experiments were performed for slow walking speeds (on average 1.0 m / sec) on a horizontal ground, and one experiment was used to represent the target ankle and torque trajectory under this walking condition. Used. The end of the stance phase was defined as the point at which the joint torque first dropped to zero after peak torque was reached in the walking cycle. This event occurred at 67% of the walking cycle in the selected experiment.

  In FIGS. 16A and 16B, the prosthetic torque and angle trajectory measured during the experiment for the subject who lost his / her foot is the biology of the subject whose body weight and height are not limb-injured. It is shown compared to the torque and angle trajectory of the ankle joint. 16A and 16B, the ankle torque (A in FIG. 16) and the ankle joint angle (B in FIG. 16) are the same from the heel contact (0% cycle) to the heel contact (100% cycle) of the same foot. Shown over one walking cycle on level ground. 16A and 16B show the average 1610, 1620 (thin line) ± 1 standard deviation (dashed line) for the measured torque and angle profile of the prosthesis under neuromuscular model control, and the same walking speed (1 m / The biomechanical values 1630 and 1640 (thick lines) of the ankle joint are plotted for one walking cycle of a subject whose body weight and height are not damaged in the limb walking in seconds). The vertical lines indicate the mean start times 1650 and 1660 of the swing phase in the walking cycle of the prosthesis, and the start times 1670 and 1680 of the swing phase of the biological ankle joint (thin alternate long and short dash lines).

ここで、T_ｍはモデルのトルク出力である、T_bioは生物学的足関節トルクである。 The model parameters are adapted to the experimental data by optimization. The parameters F _max , Gain _FB , Gain _SUPP , φ _ref and φ _max were selected for adjustment. The purpose of parameter adjustment is to best match the neuromuscular model with the biological ankle torque trajectory in a specific gait condition when the corresponding biological ankle angle trajectory is input to the model. It was to find a parameter set that could be Cost function (Cost) of optimization is given as the square error between the biological torque profile in the stance phase and the torque profile of the model given the biological ankle angle trajectory: Was defined as

Where T _m is the torque output of the model and T _bio is the biological ankle torque.

Genetic algorithm optimization was selected to do an initial search (optimum search) for optimal parameter values, and a direct search (direct search) was included to identify the optimal parameter set. Both optimization methods were implemented using Matlab's Genetic-Algorithm tool. Human walking data on horizontal ground at a selected walking speed of 1.0 m / s was used to provide a reference behavior for optimization. Table 7 shows the allowable ranges for each of the optimization parameters.

The initial population was selected by the optimization tool. Table 8 shows the parameter values obtained from the parameter optimization.

  Result of parameter optimization. To confirm the effectiveness of the optimization, optimization was performed on the final parameters using the biological ankle angle profile as input to the neuromuscular model. FIG. 17 shows a comparison between the obtained torque profile and the biological torque profile.

  FIG. 17 shows a foot of a neuromuscular model having an ankle moment profile of an intact biological ankle and an angle profile of the biological ankle as an input and having optimized parameter values. In comparison with the joint moment profile, biological ankle moment (grey line) 1710, modeled dorsiflexor component (dashed line) 1720, modeled plantar flexor component (thin line) 1730, and The moment (dashed line) 1740 of the entire neuromuscular model (plantar flexor and dorsiflexor) is shown. The ankle moment of the neuromuscular model matches the biological ankle moment almost exactly during the majority of the walking cycle.

  Low level torque control. The physical torque actually generated at the ankle during the stance phase is derived from the combined action of the longitudinal spring and the powered drive train. The stiffness of the rotating longitudinal spring is approximately linear in the operating range and is 500 N · m / radian. Using this spring constant, the vertical spring contribution is predicted and subtracted from the desired ankle torque. The remaining torque must be generated by a powered drive train.

  The performance of the powered drive train is improved by using advance compensation, friction compensation, and feed forward techniques, as shown in FIG. An experimental study of open-loop drivetrain dynamics was conducted and used to implement those improvements (M. Eilenberg, "A Neuromuscular-Model Based Control Strategy for Powered Ankle-Foot Prostheses," Master's Thesis, Massachusetts Institute of Technology, Cambridge, MA, 2009). A comparison between the output torque and the commanded torque for walking on a horizontal ground, an uphill slope, and a downhill slope is shown in FIGS. The output torque of the prosthesis was estimated using a strain gauge in the series spring for SEA torque contribution and an ankle angle-based longitudinal spring torque estimate for the torque contribution of the longitudinal spring.

  Clinical evaluation. The prosthesis was worn on the right leg of a healthy and active 75 kg leg amputee. This subject was given time to naturally adapt to walking with the prosthesis. Walking data from these experiments were recorded using a wireless link to the prosthesis. During the walking experiment on the horizontal ground, the subject was required to walk a 10 m long passage. The scheduled target walking speed was set at 1.0 m / sec to match the walking speed of the uninjured subject. The subject started walking from about 5 m from the beginning of the passage and stopped walking after passing about 3 m from the end of the passage. A marker on the ground was used to indicate the beginning and end of a 10 meter passage. A stopwatch was used to confirm the average walking speed in each experiment by showing when the subject's center of mass passed over each marker. A total of 10 tests were performed. Experiments with walking speeds within 5% of the target speed were used for processing, resulting in 45 walking cycles. The subject was then requested to walk up the 11 ° 2m long slope at his chosen speed. In 10 uphill experiments, the subject started from a level ground about 2 m from the start of the slope on the platform and stopped after about 1 m of the slope. Next, the same route was walked in the reverse direction in 12 slope down experiments.

  Data analysis. The first three and last three walking cycles in the horizontal ground experiment were ignored because they were assumed to be transient. Each of the remaining gait cycles was resampled to cover 1000 data points. The mean and standard deviation trajectories were calculated from the data obtained. The last step (one step) on the slope, both up and down the slope, was used as a representative walking cycle. Each selected walking cycle was resampled and averaged in a manner similar to that described for the level ground experiment.

  The net work in each gait cycle was calculated by numerically integrating the ankle torque over the ankle angle from heel contact to toe-off. In this case, the swing period was ignored for the net work calculation. Next, the average net work for each gait condition was calculated from the net work values for the individual gait cycles.

  result. Torque tracking. In this experiment, it was a prerequisite that the short leg brace had the ability to actually generate torque and speed commanded by the neuromuscular controller. This capability is demonstrated in FIGS. 13A-C showing commanded torque versus measured output torque for walking on a level ground, climbing a slope, and descending a slope.

Adaptation to ground inclination. The evaluation of the adaptation of the neuromuscular model controlled prosthesis to the ground slope was confirmed by the clinical trial data of FIGS. From the numerically integrated data of these tests, the net work value (work loop area) was obtained as follows.
Horizontal ground 5.4 ± 0.5 joules
Uphill slope 12.5 ± 0.6 Joule
Downhill slope 0.1 ± 1.7 Joules

   Comparison with biological knee joint. The purpose of this neuromuscular model is to represent the dynamics inherent in the human ankle-foot complex in a useful way. Thus, the controller can be evaluated based on the ability of the prosthetic controller to mimic human behavior. FIGS. 16A and 16B described above show the torque and angle profiles from the prosthesis when walking on the horizontal ground, along with those profiles of subjects with uncompromised weight and height with uninjured limbs. Show.

  18A to 18C show prosthetic devices measured for three different walking conditions: horizontal ground (A in FIG. 18), uphill slope (B in FIG. 18), and downhill slope (C in FIG. 18). Is a plot of the torque-angle trajectory. 18A to 18C show averages 1810, 1820, and 1830 ± 1 standard deviations. Arrows indicate the direction of advance with time. The average net work of the prosthesis increases as the ground slope increases. This result is shown in the document "AS McIntosh, KT Beatty, LN Dwan, and DR Vickers," Gait dynamics on an inclined walkway, "Journal of Biomechanics, Vol.39, pp 2491-2502, 2006". It is consistent with the joint data.

  For walking on level ground, the measured ankle torque and ankle angle profiles of the prosthesis qualitatively match those of an undamaged human (comparable to the prosthesis). . It may be reasonable to assume that the observed differences are subordinate and are caused by multiple factors. Factors include atrophy and / or hypertrophy of leg muscles of clinical subjects due to amputations, limb length differences, and possibly the lack of gastrocnemius muscle, a functioning biarticular muscle. Furthermore, due to the limited (detection) range of the prosthetic angle sensor, the prosthesis cannot move over the full range of motion of the undamaged ankle joint.

  Adaptation to ground inclination. The neuromuscular control presented in this specification can be uniquely adapted to the slope of the ground without clearly detecting the terrain. Increased net work of the ankle when walking uphill, and decrease net work of the ankle when walking downhill, compared to the net work of the ankle when walking on horizontal ground Are the same according to data from "AS Mclntosh, KT Beatty, LN Dwan, and DR Vickers," Gait dynamics on an inclined walkway, "Journal of Biomechanics, Vol.39, pp 2491-2502, 2006. Consistent with the behavior of an undamaged human ankle under conditions. Such changes in the positive net work during the stance phase under these walking conditions indicate the slope adaptation behavior exhibited by the neuromuscular model. The ability of these neuromuscular models to produce behavioral changes similar to these living bodies suggests that the model embodies important characteristics of human plantar flexor muscles. Furthermore, the model is also expected to have the possibility of adapting to speed. When trying to move faster, the wearer of the prosthesis may press the prosthesis more strongly. This additional force allows the modeled reflex mechanism (or reflexes) to command a stronger virtual muscle force, resulting in a greater energy output and achieving a faster walking speed.

While preferred embodiments have been disclosed, many other examples will occur to those skilled in the art, and all of these examples are within the scope of the present invention. Each of the various embodiments described above can be combined with the other described embodiments to provide many features. Furthermore, although a number of individual embodiments of the apparatus and method of the present invention have been described herein, what has been described herein is merely illustrative of the application of the principles of the present invention. Accordingly, other sequences (configurations), methods, modifications, and substitutions by one of ordinary skill in the art are also considered to be within the scope of the present invention, and the present invention is not limited except as by the appended claims.

Claims (9)

A model-based neuromachine controller for a robotic limb comprising at least one joint comprising a finite state machine, a muscle model processor, and a joint command processor
The finite state machine is configured to receive feedback data relating to the state of the robot limb and to determine the state of the robot limb using the received feedback data;
The muscle model processor receives state information from the finite state machine and receives muscle geometry and / or arrangement and reflex structure information from at least one database and uses a neuromuscular model to provide the robot Configured to determine at least one desired joint torque command or stiffness command to be sent to the limb;
The neuromechanical controller, wherein the joint command processor is configured to command biomimetic torque and stiffness determined by the muscle model processor to a joint of the robot limb.   The neuromechanical controller of claim 1, wherein the feedback data is provided by at least one sensor attached to each joint of the robot limb.   The neuromechanical controller of claim 1, wherein the robot limb is a leg and the finite state machine is synchronized to a walking cycle of the leg.   The neuromechanical controller of claim 3, wherein the leg comprises a powered short leg brace.   The neuromechanical controller of claim 3, wherein the leg comprises a knee joint.   The neuromechanical controller of claim 4, wherein the leg further comprises a knee joint.   The neuromechanical controller of claim 6, wherein the leg further comprises a hip joint.   The neuromechanical controller of claim 2, wherein the at least one sensor is a joint angular displacement and angular velocity sensor, a torque sensor, or an inertial measurement unit. A method for controlling a robotic limb comprising at least one joint comprising:
Receiving feedback data regarding the state of the robot limb with a finite state machine;
Determining the state of the robot limb using the finite state machine and the received feedback data;
At least one desired joint torque command or stiffness command to be sent to the robot limb using a neuromuscular model, muscle geometry and / or placement information, reflex structure information, and state information from the finite state machine A step of determining
Directing biomimetic torque and stiffness determined by a muscle model processor to a joint of the robot limb.

Images