WO2021110062A1 - 一种基于多传感器与拮抗式驱动的灵巧手控制系统 - Google Patents

一种基于多传感器与拮抗式驱动的灵巧手控制系统 Download PDF

Info

Publication number
WO2021110062A1
WO2021110062A1 PCT/CN2020/133437 CN2020133437W WO2021110062A1 WO 2021110062 A1 WO2021110062 A1 WO 2021110062A1 CN 2020133437 W CN2020133437 W CN 2020133437W WO 2021110062 A1 WO2021110062 A1 WO 2021110062A1
Authority
WO
WIPO (PCT)
Prior art keywords
joint
sensor
control
tendon
information
Prior art date
Application number
PCT/CN2020/133437
Other languages
English (en)
French (fr)
Inventor
任化龙
Original Assignee
深圳忆海原识科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 深圳忆海原识科技有限公司 filed Critical 深圳忆海原识科技有限公司
Publication of WO2021110062A1 publication Critical patent/WO2021110062A1/zh
Priority to US17/831,091 priority Critical patent/US20220288775A1/en

Links

Classifications

    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/16—Programme controls
    • B25J9/1628—Programme controls characterised by the control loop
    • B25J9/1633—Programme controls characterised by the control loop compliant, force, torque control, e.g. combined with position control
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
    • B25J13/081—Touching devices, e.g. pressure-sensitive
    • B25J13/084—Tactile sensors
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
    • B25J13/085—Force or torque sensors
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J13/00—Controls for manipulators
    • B25J13/08—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices
    • B25J13/087—Controls for manipulators by means of sensing devices, e.g. viewing or touching devices for sensing other physical parameters, e.g. electrical or chemical properties
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J15/00—Gripping heads and other end effectors
    • B25J15/0009—Gripping heads and other end effectors comprising multi-articulated fingers, e.g. resembling a human hand
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
    • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
    • B25J9/00—Programme-controlled manipulators
    • B25J9/10—Programme-controlled manipulators characterised by positioning means for manipulator elements
    • B25J9/104—Programme-controlled manipulators characterised by positioning means for manipulator elements with cables, chains or ribbons
    • B25J9/1045—Programme-controlled manipulators characterised by positioning means for manipulator elements with cables, chains or ribbons comprising tensioning means

Definitions

  • the present application belongs to the field of robot dexterous hand control, and specifically relates to a dexterous hand control system based on multi-sensor and antagonistic driving.
  • Bionic dexterous hand refers to a manipulator whose hand index, degree of freedom, shape and function are close to that of a human hand. It can manipulate objects flexibly and finely. It is suitable for high-performance prosthetics or used in industrial scenes such as compliant assembly. It can also replace personnel entering pollution, poisoning, Operations in hazardous environments such as radiation and service robots with strong versatility are key components of bionic robots or humanoid robots.
  • a transmission method widely used in many dexterous hand systems is tendon transmission.
  • the tendon transmission is to transmit the motion and power of the driver in the forearm to the small hand through the tendon (material is steel wire or flexible rope), and drive the related joints to rotate, which solves the problem of the small space of the dexterous hand and the difficulty of installing high-power and large-scale The contradiction of torque (or tension) drives.
  • Existing dexterous tendon transmission systems mostly adopt a scheme in which each joint is driven by a rotary drive.
  • the output shaft of the drive is fixedly connected to a winch.
  • the tendon is wound on the winch, extends to the driven pulley of the joint and winds back to the winch.
  • Human joints are driven by a pair of muscle groups in an antagonistic manner. When one group of muscles contracts and the other group of muscles is released, the corresponding joints rotate in one direction, and vice versa.
  • the two groups of muscle groups cooperate to flexibly control the joint damping and joint stiffness of the human joints, so that the human limbs can output flexible movements and strengths, which can smoothly adapt to the objects in contact and perform compliant operations, while maintaining a high level. Robustness against interference.
  • a dexterous hand using tendon transmission can also simulate this method, so that each joint is driven by a pair of drivers antagonistically, that is, when the joint is required to rotate in one direction, one driver pulls the tendon on one side, and the other driver is placed on the other side.
  • the coordination of the two drivers can flexibly control the tension of the tendons on both sides, avoid tendons slack or over-tension, and improve system reliability; this driving method can also control joint damping and joint stiffness, In turn, both compliant operation and anti-interference robustness are taken into consideration.
  • some dexterous hand systems also incorporate force feedback.
  • Some dexterous hands install a finger tip force sensor on the tip (distal knuckle) of the finger to measure the force or moment applied by the fingers of the dexterous hand to the object and the contact point; some dexterous hands have a measuring surface attached to the surface of the hand.
  • Pressure sensor array to simulate bionic skin.
  • the human hand skin has a rich sense of touch. It can perceive the contact with the object through various touch points, as well as the shape, texture and texture of the object, but cannot accurately measure the size of the contact force; there are sensors and sensors for perception and bending moments at the joints of the hand. Nerves can more accurately perceive the interaction force of each knuckle and joint when it comes in contact with an object, which is especially good for fine manipulation and further perceive the shape, texture and texture of the object with the touch of the skin; there are human tendons that can sense tendon tension.
  • the receptors and nerves can evaluate the force exerted by each finger or the entire hand on an object through the tension of the tendon, which is especially useful for estimating the force when doing pulling, pulling, and lifting heavy objects that require greater force; These structures and characteristics effectively decouple the perception processes of different levels of force and touch, and bring convenience to the comprehensive analysis of the nervous system.
  • the dexterous hand can also learn from this approach, using a variety of different sensors to sense different levels of force and tactile information, reducing the coupling between different levels of perceptual information, and facilitating flexible comprehensive analysis of multiple perceptual information under different operating tasks And simplify the control difficulty, and reduce the design complexity and cost of various types of sensors.
  • One of the purposes of the embodiments of the present application is to provide a dexterous hand control system based on multi-sensor and antagonistic driving, which aims to solve the difficulty of analyzing and controlling various sensory information of the existing dexterous hand under different operation tasks. Larger, the design of various types of sensors is complicated and the cost is higher.
  • a dexterous hand control system based on multi-sensor and antagonistic driving comprising: a dexterous hand driven by an antagonistic tendon transmission, a sensor module, a sensor management module, a driver control module, and a central control module;
  • the antagonistically driven dexterous hand driven by tendons is configured as a dexterous hand having one or more joints driven by tendon and antagonistically driven;
  • the sensor module includes a joint angle sensor set composed of multiple joint angle sensors, a tactile sensor set composed of multiple tactile sensors in the bionic skin, a joint force and torque sensor set composed of multiple joint force and torque sensors, A collection of tendon tension sensors composed of multiple tendon tension sensors;
  • the joint angle sensor is installed at each joint of the dexterous hand to measure the rotation angle of each joint, and its output signal is processed by the sensor management module to obtain joint position information;
  • the tactile sensor is distributed in the bionic skin to sense the contact with the object, and its output signal is processed by the sensor management module to obtain tactile information;
  • the joint force and moment sensors are installed at the interphalangeal joints, metacarpophalangeal joints, wrist joints, and wrist joints of the dexterous hand to measure one-to-multidimensional forces or moments at the joints, and their output signals are processed by the sensor management module Obtain joint force and moment information;
  • the tendon tension sensor is installed on the tendon to measure the tension of the tendon, and its output signal is processed by the sensor management module to obtain tendon tension information;
  • the sensor management module applies constant power to each joint angle sensor, each tactile sensor, each joint force and torque sensor, and each tendon tension sensor in the sensor module, or applies power in a periodic scanning manner, and performs processing on the output signals of these sensors. Amplify, filter, sample and convert, and monitor whether the sensor is missing or work abnormally, and transmit the processed output signal and monitoring result to the central control module;
  • the sensor management module can receive a control instruction from the central control module, and adjust a working mode according to the control instruction;
  • the drive control module has a current loop, voltage loop, and speed loop that control each drive, automatically protects the voltage or current of the drive from overload, monitors whether each drive is missing or works abnormally, and monitors the current, voltage, speed, and speed of each drive. The monitoring results are transmitted to the central control module;
  • the driver control module reads the joint angle information through the sensor management module to form a joint limit direct control loop
  • the driver control module reads joint force and torque information through the sensor management module to form a joint force and torque protection direct control loop;
  • the driver control module reads the tendon tension information through the sensor management module to form a direct control loop for tendon tension protection.
  • the driver control module can receive control instructions from the central control module, and adjust the control mode of each driver according to the control instructions, that is, the current loop, voltage loop, and Speed loop or select any combination of them to control the drive.
  • the central control module receives the operation target, controls the sensor management module and reads the information of each sensor, obtains the control signal through the process of multi-sensor information synthesis and control strategy calculation, and transmits the control signal to the drive control module to control each drive, and then to The joint position, joint speed, joint force and torque, joint damping, joint stiffness, tendon tension and contact with objects of the dexterous hand are controlled.
  • the central control module adopts a digital computer or an analog computer or an FPGA or an ASIC or a brain-like neural network chip or a combination thereof as a carrier.
  • the central control module is configured to use a deep learning neural network, a spiking neural network, and a rule-based program to perform hybrid calculations.
  • the central control module is equipped with a sensor sampling and analysis strategy, a control strategy to prevent tendon slack, a control strategy to prevent tendon overtightening, a control strategy based on controllable load, Control strategy based on dynamic model and control strategy based on neural network.
  • the control strategy based on the controllable load is to divide each driver that constitutes an antagonistic drive on the controlled joint into an active driver and a driven driver, and adjust the voltage and/or current of the driven driver to make it in a follower motion mode,
  • the driven driver is dragged by the joint through the tendon, which is equivalent to a controllable load, forming an open-loop or closed-loop control.
  • the control strategy based on the dynamic model is to establish a dynamic model of one or more drives, transmissions, joints and/or external loads that constitute an antagonistic drive, and use the dynamic model to estimate one or more state variables to form an open loop or Closed-loop control.
  • the neural network-based control strategy is to use the neural network as a controller, input one or more sensing information or state variables to the neural network, and use the output of the neural network as the control value of one or more drivers.
  • state variables include voltage, current, inertia, damping, joint position, joint speed, joint force and moment, joint damping, joint stiffness, tendon tension, load, and their functional relationship with time of one or more drives. Any one or several of them.
  • the central control module can specify the use of the control strategy based on the controllable load or the control strategy based on the dynamic model or the control strategy based on the neural network for a specific joint through configuration.
  • the control strategy of the network can specify the use of the control strategy based on the controllable load or the control strategy based on the dynamic model or the control strategy based on the neural network for a specific joint through configuration.
  • the central control module defaults to adopting the dynamic model-based control strategy or the Control strategy based on controllable load.
  • the central control module adopts the neural network-based control strategy by default; When part or all of the force and torque sensors or tendon tension sensors are missing or work abnormally, the central control module automatically switches to adopt the control strategy based on the dynamic model or the control strategy based on the controllable load to ensure the reliability of the system.
  • the sensor sampling and analysis strategy includes one or any of the following:
  • the joint force and torque sensor should be sampled first, and its sampling frequency should be increased.
  • the weight of joint force and torque information relative to other sensor information is amplified, and at the same time, it can cooperate with the information of the tactile sensor in the bionic skin to further perceive the contact between the hand and the object, as well as the shape, texture and texture of the object;
  • the joint force and torque sensors are sampled first, and the joint force and torque information is amplified relative to other sensors in the process of multi-sensor information synthesis and control strategy calculation.
  • the weight of information
  • the tendon tension sensors are sampled first, and the weight of tendon tension information relative to other sensor information is amplified in the process of multi-sensor information synthesis and control strategy calculation.
  • the sensor management module automatically adjusts the frequency of applying power to each sensor and the sampling frequency of its output signal, and its adjustment method is:
  • the power supply will be applied to the sensor
  • the frequency and the sampling frequency of its output signal are adjusted to a first preset frequency range, such as a range of 0.1 Hz to 10 Hz;
  • the frequency of applying power to the sensor and the sampling frequency of its output signal are adjusted to the first 2.
  • the preset frequency range such as the range from 10 Hz to 1000 Hz.
  • the sensor management module When the sensor management module receives a control instruction from the central control module, its working mode includes one or any of the following:
  • the joint limit direct control loop responds quickly, and directly controls the driver to limit the joint within the allowable range of motion; the central control module has forbidden joints
  • the control command of the limit direct control loop can actively prevent the joint limit direct control loop from working.
  • the central control module has a control command that prohibits the joint force and torque protection direct control loop, which can actively prevent the joint force and torque protection direct control loop from working.
  • the tendon tension protection direct control loop When the system is abnormal and the tendon tension exceeds its allowable range, the tendon tension protection direct control loop will respond quickly, and directly control the driver to limit the tendon tension within the allowable range; the central control module has a forbidden tendon
  • the tension protection direct control loop control command can actively prevent the tendon tension protection to directly control the loop operation.
  • the joint limit direct control loop, joint force and torque protection direct control loop, and tendon tension protection direct control loop make the control system more reliable.
  • the control strategy to prevent tendon relaxation is:
  • the control strategy to prevent over-tightening of tendons is:
  • the control strategy based on the controllable load is:
  • the voltage or current that the driven driver should apply is calculated through the driver model
  • the control strategy based on the dynamic model is:
  • the dynamic model includes the voltage or current, joint position, joint speed, joint force and torque of each driver that constitutes the antagonistic drive, The functional relationship between joint damping, joint stiffness, tendon tension and time; in the absence of joint force and torque sensors or tendon tension sensors, the dynamic model can be used to measure joint force and torque, joint damping, joint stiffness, and tendon tension. Make an estimate
  • the target quantity and joint position information are used as the dependent variables of the dynamic model, Solve to obtain the voltage or current to be applied by each driver constituting the antagonistic drive;
  • the joint position information and the respective voltage or current of each driver constituting the antagonistic drive are used for closed-loop control.
  • the output of the dynamic model is the estimated value of joint force and moment, joint damping, joint stiffness, and tendon tension.
  • the estimated value corresponding to the target value is used as the feedback value, and the feedback value is compared with the target value.
  • the deviation is obtained by the difference of the quantity, and the deviation is input to the control unit;
  • the joint position information and joint speed information are used as the feedback value, and the difference between the feedback value and the target value is used to obtain the deviation amount, and the deviation amount is input to the control unit ;
  • the control unit further calculates the voltage or current to be applied by each driver constituting the antagonistic drive.
  • control unit is configured to adopt a PID control rate or a neural network-based control rate.
  • the control strategy based on neural network is:
  • the output calculated by the multilayer neural network is the voltage or current to be applied by each driver.
  • the tactile sensor in the bionic skin used in the dexterous hand control system based on multi-sensor and antagonistic drive provided by the embodiments of the present application only needs to be able to sense the contact point and does not need to accurately measure the contact force
  • the joint force and torque sensor only needs to be able to sense the force or torque acting on the joint
  • the tendon tension sensor only needs to be able to make coarse-grained estimation of tendon tension, thereby reducing the design complexity and complexity of various types of sensors (and bionic skin with tactile sense).
  • the system uses a combination of various sensors and the central control module's sensor sampling and analysis strategy, which effectively decouples different levels of force and tactile sensing methods, and facilitates flexible sensing of multiple sensing information under different operating tasks Comprehensive analysis and simplified control difficulty;
  • the sensor management module can also apply power to each sensor in a periodic scanning manner, thereby reducing the power consumption and heat of each sensor, and extending the service life of the sensor;
  • the central control module of the system is equipped
  • the central control module can also be partially or completely missing or missing in the joint force and torque sensor or the tendon tension sensor. Switch to a control strategy based on a controllable load or a control strategy based on a dynamic model when the work is abnormal, so that the system can still work reliably.
  • FIG. 1 is a system block diagram of a dexterous hand control system based on multi-sensor and antagonistic drive provided by an embodiment of the application;
  • FIG. 2 is a schematic diagram of a dexterous hand joint mechanism driven by tendon transmission in a dexterous hand control system based on multi-sensor and antagonistic drive in an embodiment of the present application;
  • FIG. 3 is a schematic diagram of a wrist joint and a wrist joint of a dexterous hand in an embodiment of the application;
  • FIG. 4 is a schematic diagram of an index finger unit of a dexterous hand in an embodiment of the application
  • Fig. 5 is a block diagram of a dynamic model-based control strategy of a dexterous hand control system based on multi-sensor and antagonistic drive in an embodiment of the application.
  • an embodiment of the present application discloses a dexterous hand control system based on multi-sensor and antagonistic driving.
  • the system includes: a dexterous hand driven by an antagonistic tendon transmission, a sensor module 18, and a sensor management module 13 , Drive control module 15, Central control module 14.
  • the antagonistically driven dexterous hand driven by tendons is configured as a dexterous hand having one or more joints driven by tendons and adopting antagonistic driving.
  • the sensor module 18 includes a joint angle sensor set 9 composed of multiple joint angle sensors 19, a tactile sensor set 10 composed of multiple tactile sensors 1 in the bionic skin 7, and multiple A joint force and torque sensor assembly 11 composed of two joint force and torque sensors 3 and a tendon tension sensor assembly 12 composed of a plurality of tendon tension sensors 5.
  • the extension driver assembly 8 (which can be composed of a rotary drive output shaft connected to a winch) and the flexion drive assembly 6 (which can be a rotary drive
  • the output shaft is connected with a winch to draw the finger joint 4 through the dorsal part of the tendon and the palm side of the tendon to form an antagonistic drive:
  • extension driver assembly 8 pulls the dorsal part of the tendon, and at the same time flexes the driver assembly 6 to release the palm side part of the tendon, the finger joint 4 rotates to the dorsal side of the hand, that is, an extension movement;
  • the joint angle sensor 19 is installed at each joint of the dexterous hand to measure the rotation angle of each joint, and its output signal is processed by the sensor management module 13 to obtain joint position information; joint angle sensor 19 can adopt potentiometer type sensor or Hall sensor or optical encoder.
  • the tactile sensor 1 is distributed in the bionic skin 7 to sense contact with an object, and its output signal is processed by the sensor management module 13 to obtain tactile information.
  • the joint force and moment sensors 3 are installed at the interphalangeal joints, metacarpophalangeal joints, wrist joints, and wrist joints of the dexterous hand to measure one-to-multidimensional forces or moments at the joints, and their output signals are passed through the sensor management module 13 Process to obtain joint force and torque information; for example, at the finger joints, a preferred way is to install the joint force and torque sensor 3 at the connection between the finger joint 4 and the finger section 2, so as to effectively measure the joint force and torque at the finger joint 4.
  • FIG. 3 is a schematic diagram of a wrist joint and a palmar joint of a dexterous hand in an embodiment of the application.
  • the wrist joint refers to the connecting mechanism between the palm phalanx of each finger and the palm root of the dexterous hand.
  • the wrist joint can include multiple, such as ring finger carpal joint, little finger carpal joint, thumb carpal joint; wrist;
  • the joint is the connecting mechanism between the base of the palm of the dexterous hand and the forearm.
  • the tendon tension sensor 5 is installed on the tendon to measure the tension of the tendon, and its output signal is processed by the sensor management module 13 to obtain tendon tension information.
  • Both the joint force and moment sensor 3 and the tendon tension sensor 5 can use strain gauges as force sensitive elements.
  • the sensor management module 13 applies a constant power supply to each joint angle sensor 19, each tactile sensor 1, each joint force and torque sensor 3, and each tendon tension sensor 5 in the sensor module 18.
  • the power is applied in a periodic scanning manner, and the output signals of these sensors are amplified, filtered, sampled and converted, and whether the sensors are missing or working abnormally is monitored, and the processed output signals and monitoring results are transmitted to the central control module 14.
  • the sensor management module 13 When the joint angle sensor 19 adopts a Hall sensor or an optical encoder, the sensor management module 13 provides continuous power for it. When the joint angle sensor 19 uses a potentiometer, the sensor management module 13 provides power for it in a periodic scanning manner to save power consumption and heat.
  • the sensor management module 13 can receive control instructions from the central control module 14 and adjust the working mode according to the control instructions.
  • the sensor management module 13 may use analog-to-digital conversion devices, control devices (such as single-chip microcomputers, ARM, DSP, CPLD, FPGA), communication protocol chips, power management chips and other components to form circuits, and carry programs to realize the above functions.
  • control devices such as single-chip microcomputers, ARM, DSP, CPLD, FPGA
  • communication protocol chips such as single-chip microcomputers, ARM, DSP, CPLD, FPGA
  • power management chips and other components to form circuits, and carry programs to realize the above functions.
  • the driver control module 15 has a current loop, a voltage loop, and a speed loop that control each driver, automatically protects the voltage or current of the driver from overload, monitors whether each driver is missing or works abnormally, and monitors each driver's The current, voltage, speed and monitoring results are transmitted to the central control module 14.
  • the drive control module 15 can receive the control instructions from the central control module 14, and adjust the control mode of each drive according to the control instructions, that is, select the current loop, voltage loop, speed loop or any combination of them to control the drive. .
  • the driver control module 15 reads the joint angle information through the sensor management module 13 to form a joint limit direct control loop.
  • the driver control module 15 reads joint force and torque information through the sensor management module 13 to form a joint force and torque protection direct control loop.
  • the driver control module 15 reads tendon tension information through the sensor management module 13 to form a direct control loop for tendon tension protection.
  • the driver control module 15 may use a motor drive chip, a control device (such as a single-chip microcomputer, ARM, DSP, CPLD, FPGA), a communication protocol chip and other components to form a circuit, and carry programs to realize the above-mentioned functions.
  • a control device such as a single-chip microcomputer, ARM, DSP, CPLD, FPGA
  • a communication protocol chip and other components to form a circuit, and carry programs to realize the above-mentioned functions.
  • the central control module 14 receives the operation target, controls the sensor management module 13 and reads the information of each sensor, obtains the control signal through the process of multi-sensor information synthesis and control strategy calculation, and transmits the control signal to the drive control module 15.
  • the driver control module 15 controls the voltage and current of each driver according to the control signal, and then controls the joint position, joint speed, joint force and torque, joint damping, joint stiffness, tendon tension, and contact with objects of the dexterous hand.
  • the central control module 14 may adopt a digital computer or an analog computer or an FPGA or an ASIC or a brain-like neural network chip or a combination thereof as a carrier.
  • the central control module 14 is preferably configured to use a deep learning neural network, a pulsed neural network and a rule-based program to perform hybrid calculations.
  • the central control module 14 is equipped with sensor sampling and analysis strategies, control strategies to prevent tendon slack, control strategies to prevent tendons to be too tight, control strategies based on controllable loads, control strategies based on dynamic models, and control strategies based on neural networks. .
  • the central control module 14 can specify the control strategy based on the controllable load or the control strategy based on the dynamic model or the control strategy based on the neural network for a specific joint through configuration.
  • the central control module 14 adopts the control strategy based on the dynamic model or the control strategy based on the controllable load by default.
  • the central control module 14 adopts the neural network-based control strategy by default; when the joint force and torque sensor 3 or the tendon tension sensor 5 is partially Or when all of them are missing or work abnormally, the central control module 14 automatically switches to adopt the control strategy based on the dynamic model or the control strategy based on the controllable load to ensure the reliability of the system.
  • the moment of inertia of a single finger joint of a dexterous hand is small, and the moment of inertia is also less affected by the motion of other joints.
  • the driven driver can be simplified to have fixed inertia and controllable damping; when the driven driver is used with greater force When the tendon is traction, the equivalent damping is also greater, and vice versa; therefore, when the joint force and torque sensor 3 or the tendon tension sensor 5 is partially or completely missing or works abnormally, the central control module 14 preferably adopts the Control the load control strategy.
  • the central control module 14 preferably adopts a control strategy based on a dynamic model for control.
  • the sensor sampling and analysis strategy includes one or any of the following:
  • the tactile sensor 1 in the bionic skin 7 should be sampled first to increase its sampling frequency , And amplify the weight of tactile information relative to other sensor information in the process of multi-sensor information synthesis and control strategy calculation;
  • the joint force and torque sensor 3 should be sampled first, its sampling frequency should be increased, and the multi-sensor information should be integrated Amplifies the weight of joint force and torque information relative to other sensor information during the calculation of the control strategy. At the same time, it can cooperate with the information of the tactile sensor 1 in the bionic skin 7 to further perceive the contact between the hand and the object, as well as the shape and texture of the object. And texture.
  • FIG. 4 is a schematic diagram of the index finger unit of a dexterous hand in an embodiment of this application.
  • “knuckle” refers to a rod-shaped component
  • “joint” refers to two or more The rod-shaped parts form a connected part through a kinematic pair (generally a rotating pair).
  • the joint force and torque sensors 3 are sampled first, and the joint force and torque information is amplified in the process of multi-sensor information synthesis and control strategy calculation.
  • the weight of other sensor information
  • the tendon tension sensor 5 When part or all of the joint force and torque sensors 3 are missing or fail, the tendon tension sensor 5 will be sampled first, and the tendon tension information will be amplified relative to other sensor information in the process of multi-sensor information synthesis and control strategy calculation. Weights.
  • the sensor management module 13 automatically adjusts the frequency of applying power to each sensor and the sampling frequency of its output signal.
  • the adjustment method is as follows:
  • the frequency of applying power to the sensor and the sampling frequency of its output signal are adjusted to the first 2.
  • the preset frequency range such as the range from 10 Hz to 1000 Hz.
  • its working mode includes one or any of the following:
  • the joint limit direct control loop responds quickly, and directly controls the driver to limit the joint to the allowable range of motion; the central control module 14 It has a control command that prohibits the joint limit direct control loop, which can actively prevent the joint limit direct control loop from working.
  • the joint force and torque protection direct control loop will respond quickly, and directly control the driver to limit the joint force and torque within the allowable range.
  • the central control module 14 has a control command that prohibits the joint force and torque protection direct control loop, which can actively prevent the joint force and torque protection direct control loop from working.
  • the tendon tension protection direct control loop will respond quickly and directly control the driver to limit the tendon tension within the allowable range; the central control module 14 It has a control command to prohibit the direct control loop of tendon tension protection, which can actively prevent the direct control loop of tendon tension protection from working.
  • the joint limit direct control loop, joint force and torque protection direct control loop, and tendon tension protection direct control loop make the control system more reliable.
  • the control strategy to prevent tendon relaxation is:
  • the control strategy to prevent over-tightening of tendons is:
  • the movement posture of the dexterous hand will also have certain errors. Based on this, those skilled in the art usually make a dexterous hand work as long as it meets a certain error range.
  • the above-mentioned "slightly higher” means to make the tendon tension higher than the lowest threshold in a smaller range, such as the first range (0.5 to 1N); and “slightly lower” means to make the tendon tension lower than the highest threshold by a smaller range.
  • the control strategy based on the controllable load is:
  • the voltage or current that the driven driver should apply is calculated through the driver model
  • control strategy based on the dynamic model is:
  • the dynamic model 17 includes the voltage or current, joint position, joint speed, joint force, and joint force of each driver that constitutes the antagonistic drive.
  • the target quantity and joint position information are used as the dependent variables of the dynamic model 17 , Solve to obtain the voltage or current (or is it a function of the voltage or current change with time) that each driver that constitutes the antagonistic drive should be applied;
  • the joint position information and the respective voltage or current of each driver constituting the antagonistic drive are used for closed-loop control.
  • the output of the dynamic model 17 is the estimated value of joint force and moment, joint damping, joint stiffness, and tendon tension.
  • the estimated value corresponding to the target amount is used as the feedback amount, and the feedback amount Make the difference with the target amount to obtain the deviation amount, and input the deviation amount to the control unit 16;
  • the joint position information and joint speed information are used as the feedback value, and the difference between the feedback value and the target value is used to obtain the deviation amount, and the deviation amount is input to the control unit 16;
  • the control unit 16 further calculates the voltage or current to be applied by each driver constituting the antagonistic drive.
  • the control unit 16 may adopt a PID control rate or a neural network-based control rate.
  • the control strategy based on neural network is:
  • the multi-layer neural network can adopt a spiking neural network that simulates the basal nucleus and cerebellum of the biological brain;
  • the output calculated by the multilayer neural network is the voltage or current to be applied by each driver.

Abstract

一种基于多传感器与拮抗式驱动的灵巧手控制系统，包括由腱传动的拮抗式驱动的灵巧手、关节角度传感器（19）、仿生皮肤中的触觉传感器（1）、关节力和力矩传感器（3）、腱张力传感器（5）、传感器管理模块（13）、驱动器控制模块（15）、中央控制模块（14）；系统将不同层次的力和触觉的感知方式有效解耦，能在不同操作任务下对多种感知信息进行灵活的综合分析，简化了控制难度，降低了各类型传感器以及具有触觉的仿生皮肤的设计复杂度和成本；系统的中央控制模块（14）使灵巧手能够避免腱过度松弛或张紧，能够控制各个关节的关节阻尼和关节刚度，兼顾柔顺操作与抗干扰鲁棒性，还能够使系统在关节力和力矩传感器（3）或腱张力传感器（5）部分或全部缺失或工作异常时仍能可靠工作。

Description

一种基于多传感器与拮抗式驱动的灵巧手控制系统
本申请要求于2019年12月02日在中国专利局提交的、申请号为201911213420.7的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请属于机器人灵巧手的控制领域，具体涉及一种基于多传感器与拮抗式驱动的灵巧手控制系统。
 背景技术
仿生灵巧手是指手指数、自由度、形状和功能接近人手的机械手，能够灵活精细地操作物体，适合作为高性能假肢，或用于柔顺装配等工业场景，还可以代替人员进入污染、毒害、辐射等危险环境作业，以及应用于通用性较强的服务型机器人，是仿生机器人或人形机器人的关键组成部分。
目前一种被广泛应用于诸多灵巧手系统的传动方式是腱传动。腱传动是通过腱（材质为钢丝或柔绳）将位于小臂中的驱动器的运动和动力传递至狭小的手部，带动相关关节转动，解决了灵巧手手部空间狭小不易安装大功率、大扭矩（或拉力）驱动器的矛盾。既有灵巧手的腱传动系统大多采用每个关节由一个旋转型驱动器驱动的方案，驱动器输出轴固联一个绞盘，腱缠绕在绞盘上，延伸至关节的从动滑轮并缠绕后回到绞盘从而形成闭环；当驱动器旋转时，绞盘将其缠绕的腱一侧收，另一侧放，从而带动关节转动。但腱的传输路径中存在间隙，且容易受到其它途径的关节的影响，这种方式很难保证两侧的腱的收缩和释放长度始终保持一致，导致两侧的腱时而发生松弛或过度绷紧，不够可靠。同时，腱松弛还会带来控制方面的问题。
人的关节由一对肌肉群组形成拮抗式驱动，当一组肌肉收缩，另一组肌肉释放，对应关节即向一个方向转动，反之亦然。两组肌肉群配合，可以灵活控制人体关节的关节阻尼和关节刚度，使人的肢体能够输出灵活变化的运动和力道，既能够柔顺地适应接触的物体以及进行柔顺操作，同时又能够保持较高的抗干扰鲁棒性。采用腱传动的灵巧手也可模拟此方式，令每个关节由一对驱动器进行拮抗式驱动，即需要关节向一个方向转动时，一个驱动器拉一侧的腱，另一个驱动器放另一侧的腱，反之亦然；两个驱动器协调配合，可以灵活地控制两侧腱的张紧程度，避免腱松弛或过度张紧，提高系统可靠性；这种驱动方式还可以控制关节阻尼和关节刚度，进而兼顾柔顺操作和抗干扰鲁棒性。
除采用位置反馈外，一些灵巧手系统还融入了力反馈。有的灵巧手在手指的尖端（远指节）安装指端力传感器，用于测量灵巧手手指对物体施加的力或力矩以及接触点；还有的灵巧手在手部表面表贴了具有测量压力的传感器阵列以模拟仿生皮肤。这些方案往往要求传感器既能够感知触点又能够精确测量力的大小，使传感器的设计和生产变得复杂，应用成本也较高。
人手皮肤具有丰富的触感，能够通过各个触点感知与物体的接触情况，以及感知物体的形状、纹理和质地，而不能精确测量接触力的大小；人手关节处存在感知力和弯矩的感受器和神经，可以较精确地感知各个指节和关节与物体接触时的相互作用力，尤其利于精细操作以及配合皮肤的触觉进一步感知物体的形状、纹理和质地；人的肌腱中存在能够感知肌腱张力的感受器和神经，可以通过肌腱张力评估各个手指或手部整体对物体施加的力，尤其利于在做拉扯、扳扣以及拎提重物等需要较大力道的动作时对力进行估测；人手的这些结构和特点将不同层次的力和触觉的感知过程有效解耦，为神经系统的综合分析带来了便利。灵巧手也可借鉴这种方式，采用多种不同的传感器分别感知不同层次的力和触觉信息，减少不同层次感知信息间的耦合，便于在不同操作任务下对多种感知信息进行灵活的综合分析以及简化控制难度，并降低各类型传感器的设计复杂度和成本。
 技术问题
本申请实施例的目的之一在于：提供一种基于多传感器与拮抗式驱动的灵巧手控制系统，旨在解决现有的灵巧手在不同操作任务下对多种感知信息进行分析以及控制时难度较大，各类型传感器的设计复杂、成本较高的问题。
 技术解决方案
为解决上述技术问题，本申请实施例采用的技术方案是：
一种基于多传感器与拮抗式驱动的灵巧手控制系统，所述系统包括：由腱传动的拮抗式驱动的灵巧手、传感器模块、传感器管理模块、驱动器控制模块、中央控制模块；
所述由腱传动的拮抗式驱动的灵巧手配置为具有一至多个由腱传动并且采用拮抗式驱动的关节的灵巧手；
所述传感器模块包括由多个关节角度传感器构成的关节角度传感器集合、由仿生皮肤中的多个触觉传感器构成的触觉传感器集合、由多个关节力和力矩传感器构成的关节力和力矩传感器集合、由多个腱张力传感器构成的腱张力传感器集合；
所述关节角度传感器安装于灵巧手的各个关节处，测量各个关节的旋转角度，其输出信号经所述传感器管理模块处理得到关节位置信息；
所述触觉传感器分布于仿生皮肤中，感知与物体的接触情况，其输出信号经所述传感器管理模块处理得到触觉信息；
所述关节力和力矩传感器安装于灵巧手的各个指间关节、掌指关节、腕掌关节、腕部关节处，测量关节处的一至多维力或力矩，其输出信号经所述传感器管理模块处理得到关节力和力矩信息；
所述腱张力传感器安装在腱上，测量腱的张力，其输出信号经所述传感器管理模块处理得到腱张力信息；
所述传感器管理模块对传感器模块中的各个关节角度传感器、各个触觉传感器、各个关节力和力矩传感器、各个腱张力传感器施加恒定电源或以周期扫描的方式施加电源，并对这些传感器的输出信号进行放大、滤波、采样与转换，以及监测是否有传感器缺失或工作异常，并将处理过的输出信号与监测结果传给中央控制模块；
在本申请实施例的一种可能的实现方式中，所述传感器管理模块能够接收中央控制模块传来的控制指令，并根据控制指令调整工作方式；
所述驱动器控制模块具有控制各个驱动器的电流环、电压环、速度环，自动对驱动器的电压或电流进行过载保护，监测各个驱动器是否缺失或工作异常，并将各个驱动器的电流、电压、速度和监测结果传给中央控制模块；
所述驱动器控制模块通过所述传感器管理模块读取关节角度信息，构成关节限位直接控制环路；
所述驱动器控制模块通过所述传感器管理模块读取关节力和力矩信息，构成关节力和力矩保护直接控制环路；
所述驱动器控制模块通过所述传感器管理模块读取腱张力信息，构成腱张力保护直接控制环路。
在本申请实施例的一种可能的实现方式中，所述驱动器控制模块能够接收中央控制模块传来的控制指令，并根据控制指令调整各个驱动器的控制方式，即单独选择电流环、电压环、速度环或选择它们的任意组合对驱动器进行控制。
所述中央控制模块接收操作目标，控制传感器管理模块并读取各个传感器的信息，经过多传感器信息综合与控制策略计算过程得到控制信号，将控制信号传给驱动器控制模块以控制各个驱动器，进而对灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及与物体的接触情况进行控制。
在本申请实施例的一种可能的实现方式中，所述中央控制模块采用数字计算机或模拟计算机或FPGA或ASIC或类脑神经网络芯片或它们的组合为载体。
在本申请实施例的一种可能的实现方式中，所述中央控制模块配置为采用深度学习神经网络、脉冲神经网络和基于规则的程序进行混合计算。
在本申请实施例的一种可能的实现方式中，所述中央控制模块搭载有传感器采样与分析策略、防止腱松弛的控制策略、防止腱过紧的控制策略、基于可控负载的控制策略、基于动态模型的控制策略以及基于神经网络的控制策略。
所述基于可控负载的控制策略为，将对受控关节构成拮抗式驱动的各个驱动器分为主动驱动器和从动驱动器，通过调整从动驱动器的电压和/或电流使其处于跟随运动模式，使从动驱动器通过腱被关节拖动从而等效为可控负载，形成开环或闭环控制。
所述基于动态模型的控制策略为，建立构成拮抗式驱动的一至多个驱动器、传动装置、关节和/或外部负载的动态模型，通过动态模型对一至多个状态变量进行估计，形成开环或闭环控制。
所述基于神经网络的控制策略为，采用神经网络作为控制器，将一至多个传感信息或状态变量输入至所述神经网络，将所述神经网络的输出作为一至多个驱动器的控制量。
其中，所述状态变量包括一至多个驱动器的电压、电流、惯性、阻尼、关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、负载以及它们与时间的函数关系等中的任一种或任几种。
在本申请实施例的一种可能的实现方式中，所述中央控制模块能够通过配置指定对特定关节采用所述基于可控负载的控制策略或所述基于动态模型的控制策略或所述基于神经网络的控制策略。
在本申请实施例的一种可能的实现方式中，在灵巧手没有安装关节力和力矩传感器或腱张力传感器的情况下，所述中央控制模块默认采用所述基于动态模型的控制策略或所述基于可控负载的控制策略。
在本申请实施例的一种可能的实现方式中，在灵巧手安装了关节力和力矩传感器以及腱张力传感器的情况下，所述中央控制模块默认采用所述基于神经网络的控制策略；当关节力和力矩传感器或腱张力传感器部分或全部缺失或工作异常时，所述中央控制模块自动切换为采用所述基于动态模型的控制策略或所述基于可控负载的控制策略，以保证系统可靠。
所述传感器采样与分析策略包括下面的一种或任几种：
1）在需要感知和物体的接触情况，或需要感知物体的形状、纹理和质地，而不需要精确测量接触力的大小时，优先对仿生皮肤中的触觉传感器进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大触觉信息相对其它传感信息的权重；
2）在需要精细操作并需要精确感知手部各个指节和关节与物体接触的相互作用力和力矩时，优先对关节力和力矩传感器进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重，同时能够配合仿生皮肤中的触觉传感器的信息进一步感知手部和物体的接触情况，以及感知物体的形状、纹理和质地；
3）在做拉扯、扳扣以及拎提重物等需要预设力道的动作时，优先对腱张力传感器进行采样，提高其采样频率，以估测各个手指或手部整体对物体施加的力，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重，同时能够配合关节力和力矩传感器的信息进一步对关节力或力矩进行精确测量；
4）在仿生皮肤的部分或全部触觉传感器缺失或失效的情况下，优先对关节力和力矩传感器进行采样，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重；
5）在部分或全部关节力和力矩传感器缺失或失效的情况下，优先对腱张力传感器进行采样，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重。
所述传感器管理模块对各个传感器施加电源的频率和对其输出信号的采样频率进行自动调节，其调节方式为：
1）当某个传感器的输出信号幅度在一定时间范围内持续低于敏感阈值，或输出信号幅度随时间的变化率在一定时间范围内持续低于变化率阈值，则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第一预设频率范围，如0.1Hz到10HZ的范围；
2）当某个传感器的输出信号幅度高于敏感阈值，或输出信号幅度随时间的变化率高于变化率阈值，则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第二预设频率范围，如10Hz到1000Hz的范围。
所述传感器管理模块在接收到中央控制模块的控制指令时，其工作方式包括下面的一种或任几种：
1）针对控制指令指定的传感器以指定的频率施加电源；
2）针对控制指令指定的传感器进行监测；
3）以控制指令指定的倍数放大指定的传感器的输出信号；
4）以控制指令指定的滤波方式对指定的传感器的输出信号进行滤波；
5）优先对控制指令指定的传感器的输出信号进行采样与转换；
6）以控制指令指定的采样频率对指定的传感器的输出信号进行采样与转换；
当系统发生异常，使关节超出了其容许的运动范围时，所述关节限位直接控制环路进行快速响应，直接控制驱动器将关节限制在容许的运动范围内；所述中央控制模块具有禁止关节限位直接控制环路的控制指令，能够主动阻止关节限位直接控制环路工作。
当系统发生异常，使关节力和力矩超出了其容许的范围时，所述关节力和力矩保护直接控制环路进行快速响应，直接控制驱动器将关节力和力矩限制在容许的范围内；所述中央控制模块具有禁止关节力和力矩保护直接控制环路的控制指令，能够主动阻止关节力和力矩保护直接控制环路工作。
当系统发生异常，使腱张力超出了其容许的范围时，所述腱张力保护直接控制环路进行快速响应，直接控制驱动器将腱张力限制在容许的范围内；所述中央控制模块具有禁止腱张力保护直接控制环路的控制指令，能够主动阻止腱张力保护直接控制环路工作。
所述关节限位直接控制环路、关节力和力矩保护直接控制环路、腱张力保护直接控制环路使该控制系统更加可靠。
所述防止腱松弛的控制策略为：
1）设定腱张力的最低阈值；
2）通过腱张力传感器判断腱张力是否低于最低阈值；
3）若腱张力低于最低阈值，则使牵引该腱的驱动器进一步收紧腱，直至腱张力等于或略高于最低阈值。
所述防止腱过紧的控制策略为：
1）设定腱张力的最高阈值；
2）通过腱张力传感器判断腱张力是否高于最高阈值；
3）若腱张力高于最高阈值，则使牵引该腱的驱动器进一步释放腱，直至腱张力等于或略低于最高阈值。
所述基于可控负载的控制策略为：
1）将对受控关节构成拮抗式驱动的各个驱动器分为主动驱动器和从动驱动器；
2）通过调整从动驱动器的电压或电流使其处于跟随运动模式，使从动驱动器通过腱被关节拖动从而等效为具有惯性和能够控制阻尼的可控负载；
3）根据关节速度信息以及可控负载所需产生的阻尼大小，通过驱动器的模型计算得到从动驱动器应施加的电压或电流；
4）将主动驱动器通过腱拖动关节进而拖动可控负载的运动过程作为控制对象；该控制对象等效为单驱动器驱动的机器人关节运动控制过程，因此将控制问题有效地简化了；
5）将由灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量，将关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量，形成开环或闭环的控制。
所述基于动态模型的控制策略为：
1）建立构成拮抗式驱动的各个驱动器、传动装置、关节和外部负载的动态模型；所述动态模型包括构成拮抗式驱动的各个驱动器的电压或电流、关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及时间之间的函数关系；在缺少关节力和力矩传感器或腱张力传感器的情况下能够通过所述动态模型对关节力和力矩、关节阻尼、关节刚度、腱张力进行估计；
2）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行开环控制时，将目标量和关节位置信息作为所述动态模型的因变量，求解得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流；
3）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行闭环控制时，将关节位置信息、构成拮抗式驱动的各个驱动器各自的电压或电流作为所述动态模型的输入，所述动态模型的输出即为关节力和力矩、关节阻尼、关节刚度、腱张力的估计值，将目标量对应的估计值作为反馈量，并将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元；
4）当对目标关节位置、目标关节速度作为目标量进行闭环控制时，将关节位置信息和关节速度信息作为反馈量，将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元；
5）经控制单元进一步计算得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流。
在本申请实施例的一种可能的实现方式中，所述控制单元配置为采用PID控制率或基于神经网络的控制率。
所述基于神经网络的控制策略为：
1）构建多层神经网络作为运动控制器；
2）将由灵巧手的一至多个关节的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量输入给所述多层神经网络；
3）将由灵巧手的一至多个关节的关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量输入给所述多层神经网络；
4）经过所述多层神经网络计算得到的输出即为各个驱动器分别应施加的电压或电流。
 有益效果
本申请实施例的有益效果为：本申请实施例提供的基于多传感器与拮抗式驱动的灵巧手控制系统所采用的仿生皮肤中的触觉传感器仅需要能够感知接触点而不需要精确测量接触力，关节力和力矩传感器仅需要能够感知作用在关节的力或力矩，腱张力传感器仅需要能够对腱张力进行粗粒度估测，从而降低各类型传感器（以及具有触觉的仿生皮肤）的设计复杂度和成本；该系统采用的各种传感器的组合以及中央控制模块的传感器采样与分析策略，将不同层次的力和触觉的感知方式有效解耦，便于在不同操作任务下对多种感知信息进行灵活的综合分析以及简化控制难度；所述传感器管理模块还能够对各个传感器以周期扫描的方式施加电源从而减少了各个传感器的功耗与发热，并且延长了传感器的使用寿命；该系统的中央控制模块搭载有防止腱松弛的控制策略、防止腱过紧的控制策略、基于可控负载的控制策略、基于动态模型的控制策略以及基于神经网络的控制策略，这些策略能够有效避免腱过度松弛或张紧，并且能够控制灵巧手各个关节的关节阻尼和关节刚度，使灵巧手能够兼顾柔顺操作与抗干扰鲁棒性；所述中央控制模块还能够在关节力和力矩传感器或腱张力传感器部分或全部缺失或工作异常时切换至基于可控负载的控制策略或基于动态模型的控制策略，使系统仍能可靠工作。
 附图说明
为了更清楚地说明本申请实施例中的技术方案，下面将对实施例或示范性技术描述中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本申请的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1为本申请实施例提供的一种基于多传感器与拮抗式驱动的灵巧手控制系统的系统框图；
图2为本申请实施例中一种基于多传感器与拮抗式驱动的灵巧手控制系统的由腱传动的拮抗式驱动的灵巧手关节机构示意图；
图3为本申请实施例中一种灵巧手的腕部关节与腕掌关节的示意图；
图4为本申请实施例中一种灵巧手的食指单元的示意图；
图5为本申请实施例中一种基于多传感器与拮抗式驱动的灵巧手控制系统的基于动态模型的控制策略的框图。
 本发明的实施方式
为了使本申请的目的、技术方案及优点更加清楚明白，以下结合附图及实施例，对本申请进行详细说明。应当理解，此处所描述的具体实施例仅用以解释本申请，并不用于限定本申请。
为了说明本申请的技术方案，以下结合具体附图及实施例进行详细说明。
参见附图1，本申请实施例公开了一种基于多传感器与拮抗式驱动的灵巧手控制系统，所述系统包括：由腱传动的拮抗式驱动的灵巧手、传感器模块18、传感器管理模块13、驱动器控制模块15、中央控制模块14。
所述由腱传动的拮抗式驱动的灵巧手配置为具有一至多个由腱传动并且采用拮抗式驱动的关节的灵巧手。
参见附图1和附图2，所述传感器模块18包括由多个关节角度传感器19构成的关节角度传感器集合9、由仿生皮肤7中的多个触觉传感器1构成的触觉传感器集合10、由多个关节力和力矩传感器3构成的关节力和力矩传感器集合11、由多个腱张力传感器5构成的腱张力传感器集合12。
参见附图2，在一个由腱传动并且采用拮抗式驱动的灵巧手关节机构中，伸展驱动器组件8（可由一个旋转型驱动器输出轴联接一个绞盘构成）和屈曲驱动器组件6（可由一个旋转型驱动器输出轴联接一个绞盘构成）分别通过腱的手背侧部分和腱的手心侧部分牵引手指关节4，构成拮抗式驱动：
1）当伸展驱动器组件8牵拉腱的手背侧部分，同时屈曲驱动器组件6释放腱的手心侧部分，手指关节4向手背侧旋转，即伸展运动；
2）当伸展驱动器组件8释放腱的手背侧部分，同时屈曲驱动器组件6牵拉腱的手心侧部分，手指关节4向手心侧旋转，即屈曲运动。
参见附图1和附图2，所述关节角度传感器19安装于灵巧手的各个关节处，测量各个关节的旋转角度，其输出信号经所述传感器管理模块13处理得到关节位置信息；关节角度传感器19可以采用电位计式传感器或霍尔传感器或光编码器。
所述触觉传感器1分布于仿生皮肤7中，感知与物体的接触情况，其输出信号经所述传感器管理模块13处理得到触觉信息。
所述关节力和力矩传感器3安装于灵巧手的各个指间关节、掌指关节、腕掌关节、腕部关节处，测量关节处的一至多维力或力矩，其输出信号经所述传感器管理模块13处理得到关节力和力矩信息；例如，在手指关节处，一种优选的方式是将关节力和力矩传感器3安装于手指关节4与手指节2的连接处，以有效测量手指关节4处的一至多维力和力矩。参见附图3，为本申请实施例中一种灵巧手的腕部关节与腕掌关节的示意图。其中，腕掌关节是指灵巧手中每个手指掌指节与手掌根部之间的联接机构，腕掌关节可以包括多个，如无名指腕掌关节、小指腕掌关节、拇指腕掌关节；腕部关节则是灵巧手手掌根部与小臂之间的连接机构。
所述腱张力传感器5安装在腱上，测量腱的张力，其输出信号经所述传感器管理模块13处理得到腱张力信息。
所述关节力和力矩传感器3和腱张力传感器5均可采用应变片为力敏感元件。
参见附图1和附图2，所述传感器管理模块13对传感器模块18中的各个关节角度传感器19、各个触觉传感器1、各个关节力和力矩传感器3、各个腱张力传感器5施加恒定电源或以周期扫描的方式施加电源，并对这些传感器的输出信号进行放大、滤波、采样与转换，以及监测是否有传感器缺失或工作异常，并将处理过的输出信号与监测结果传给中央控制模块14。
当关节角度传感器19采用霍尔型传感器或光编码器时，所述传感器管理模块13为其提供持续电源。当关节角度传感器19采用电位计时，所述传感器管理模块13以周期扫描的方式为其提供电源，以节省功耗和发热。
所述传感器管理模块13可以接收中央控制模块14传来的控制指令，并根据控制指令调整工作方式。
所述传感器管理模块13可以采用模数转换器件、控制器件（如单片机、ARM、DSP、CPLD、FPGA）、通讯协议芯片、电源管理芯片等元件构成电路，并搭载程序以实现上述功能。
参见附图1，所述驱动器控制模块15具有控制各个驱动器的电流环、电压环、速度环，自动对驱动器的电压或电流进行过载保护，监测各个驱动器是否缺失或工作异常，并将各个驱动器的电流、电压、速度和监测结果传给中央控制模块14。
所述驱动器控制模块15可以接收中央控制模块14传来的控制指令，并根据控制指令调整各个驱动器的控制方式，即单独选择电流环、电压环、速度环或选择它们的任意组合对驱动器进行控制。
所述驱动器控制模块15通过所述传感器管理模块13读取关节角度信息，构成关节限位直接控制环路。
所述驱动器控制模块15通过所述传感器管理模块13读取关节力和力矩信息，构成关节力和力矩保护直接控制环路。
所述驱动器控制模块15通过所述传感器管理模块13读取腱张力信息，构成腱张力保护直接控制环路。
所述驱动器控制模块15可以采用电机驱动芯片、控制器件（如单片机、ARM、DSP、CPLD、FPGA）、通讯协议芯片等元件构成电路，并搭载程序以实现上述功能。
参见附图1，所述中央控制模块14接收操作目标，控制传感器管理模块13并读取各个传感器的信息，经过多传感器信息综合与控制策略计算过程得到控制信号，将控制信号传给驱动器控制模块15，驱动器控制模块15根据控制信号控制各个驱动器的电压和电流，进而对灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及与物体的接触情况进行控制。
所述中央控制模块14可采用数字计算机或模拟计算机或FPGA或ASIC或类脑神经网络芯片或它们的组合为载体。
所述中央控制模块14优选配置为采用深度学习神经网络、脉冲神经网络和基于规则的程序进行混合计算。
所述中央控制模块14搭载有传感器采样与分析策略、防止腱松弛的控制策略、防止腱过紧的控制策略、基于可控负载的控制策略、基于动态模型的控制策略以及基于神经网络的控制策略。
所述中央控制模块14可通过配置指定对特定关节采用所述基于可控负载的控制策略或所述基于动态模型的控制策略或所述基于神经网络的控制策略。
在灵巧手没有安装关节力和力矩传感器3或腱张力传感器5的情况下，所述中央控制模块14默认采用所述基于动态模型的控制策略或所述基于可控负载的控制策略。
在灵巧手安装了关节力和力矩传感器3以及腱张力传感器5的情况下，所述中央控制模块14默认采用所述基于神经网络的控制策略；当关节力和力矩传感器3或腱张力传感器5部分或全部缺失或工作异常时，所述中央控制模块14自动切换为采用所述基于动态模型的控制策略或所述基于可控负载的控制策略，以保证系统可靠。
灵巧手单个手指关节的转动惯量较小，且转动惯量受其它关节运动的影响也较小，可将从动驱动器简化为具有固定的惯性和可控的阻尼；当使从动驱动器以较大力量牵引腱时，其等效的阻尼也较大，反之亦然；因此，手指关节在关节力和力矩传感器3或腱张力传感器5部分或全部缺失或工作异常时，中央控制模块14优选采用基于可控负载的控制策略进行控制。
灵巧手手部整体的转动惯量较大，且转动惯量随手部姿态变化而变化剧烈，需要较复杂的模型进行计算，因此腕部关节在关节力和力矩传感器3或腱张力传感器5部分或全部缺失或工作异常时，中央控制模块14优选采用基于动态模型的控制策略进行控制。
所述传感器采样与分析策略包括下面的一种或任几种：
1）在需要感知和物体的接触情况，或需要感知物体的形状、纹理和质地，而不需要精确测量接触力的大小时，优先对仿生皮肤7中的触觉传感器1进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大触觉信息相对其它传感信息的权重；
2）在需要精细操作并需要精确感知手部各个指节和关节与物体接触的相互作用力和力矩时，优先对关节力和力矩传感器3进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重，同时能够配合仿生皮肤7中的触觉传感器1的信息进一步感知手部和物体的接触情况，以及感知物体的形状、纹理和质地。参见附图4，为本申请实施例中一种灵巧手的食指单元的示意图，在机械手的常见概念中，“指节”指的是杆状部件，而“关节”指的是两或多个杆状部件通过运动副（一般是转动副）形成联接的部分。
3）在做拉扯、扳扣以及拎提重物等需要预设力道（即较大的力道范围，例如指端力需要输出至少50N）的动作时，优先对腱张力传感器5进行采样，提高其采样频率，以估测各个手指或手部整体对物体施加的力，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重，同时能够配合关节力和力矩传感器3的信息进一步对关节力或力矩进行精确测量；
4）在仿生皮肤7的部分或全部触觉传感器1缺失或失效的情况下，优先对关节力和力矩传感器3进行采样，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重；
5）在部分或全部关节力和力矩传感器3缺失或失效的情况下，优先对腱张力传感器5进行采样，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重。
所述传感器管理模块13对各个传感器施加电源的频率和对其输出信号的采样频率进行自动调节，其调节方式为：
1）当某个传感器的输出信号幅度在一定时间范围（如10秒）内持续低于敏感阈值（如：将关节力和力矩传感器3的敏感阈值设为0.2牛），或输出信号幅度随时间的变化率在一定时间范围（如10秒）内持续低于变化率阈值（如：将关节力和力矩传感器3的变化率阈值设为正负0.1牛/秒），则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第一预设频率范围，如0.1Hz到10HZ的范围；
2）当某个传感器的输出信号幅度高于敏感阈值，或输出信号幅度随时间的变化率高于变化率阈值，则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第二预设频率范围，如10Hz到1000Hz的范围。
所述传感器管理模块13在接收到中央控制模块14的控制指令时，其工作方式包括下面的一种或任几种：
1）针对控制指令指定的传感器以指定的频率施加电源；
2）针对控制指令指定的传感器进行监测；
3）以控制指令指定的倍数放大指定的传感器的输出信号；
4）以控制指令指定的滤波方式对指定的传感器的输出信号进行滤波；
5）优先对控制指令指定的传感器的输出信号进行采样与转换；
6）以控制指令指定的采样频率对指定的传感器的输出信号进行采样与转换。
参见附图1，当系统发生异常，使关节超出了其容许的运动范围时，关节限位直接控制环路进行快速响应，直接控制驱动器将关节限制在容许的运动范围内；所述中央控制模块14具有禁止关节限位直接控制环路的控制指令，能够主动阻止关节限位直接控制环路工作。
参见附图1，当系统发生异常，使关节力和力矩超出了其容许的范围时，关节力和力矩保护直接控制环路进行快速响应，直接控制驱动器将关节力和力矩限制在容许的范围内；所述中央控制模块14具有禁止关节力和力矩保护直接控制环路的控制指令，能够主动阻止关节力和力矩保护直接控制环路工作。
参见附图1，当系统发生异常，使腱张力超出了其容许的范围时，腱张力保护直接控制环路进行快速响应，直接控制驱动器将腱张力限制在容许的范围内；所述中央控制模块14具有禁止腱张力保护直接控制环路的控制指令，能够主动阻止腱张力保护直接控制环路工作。
所述关节限位直接控制环路、关节力和力矩保护直接控制环路、腱张力保护直接控制环路使该控制系统更加可靠。
所述防止腱松弛的控制策略为：
1）设定腱张力的最低阈值；
2）通过腱张力传感器5判断腱张力是否低于最低阈值；
3）若腱张力低于最低阈值，则使牵引该腱的驱动器进一步收紧腱，直至腱张力等于或略高于最低阈值。
所述防止腱过紧的控制策略为：
1）设定腱张力的最高阈值；
2）通过腱张力传感器5判断腱张力是否高于最高阈值；
3）若腱张力高于最高阈值，则使牵引该腱的驱动器进一步释放腱，直至腱张力等于或略低于最高阈值。
需要说明的是，由于灵巧手中各个关节、指节、传动部件的形状、造型、以及它们之间的配合安装会存在一定的误差，则灵巧手的运动姿态也会存在一定误差。基于此，本领域的技术人员通常使得灵巧手只要满足某个误差范围即可工作。上述“略高于”是指使腱张力高于所述最低阈值一个较小的范围，如第一范围（0.5至1N）；而“略低于”是指使腱张力低于所述最高阈值一个较小的范围，如第二范围（0.5至1N）。实际上，这两个范围都是能够灵活调整的，可大可小。即，上述“略高于”、“略低于”是基于灵巧手在运动姿态中不可避免和克服的误差，而提出的一个状态值。
所述基于可控负载的控制策略为：
1）将对受控关节构成拮抗式驱动的各个驱动器分为主动驱动器和从动驱动器；
2）通过调整从动驱动器的电压或电流使其处于跟随运动模式，使从动驱动器通过腱被关节拖动从而等效为具有惯性和能够控制阻尼的可控负载；
3）根据关节速度信息以及可控负载所需产生的阻尼大小，通过驱动器的模型计算得到从动驱动器应施加的电压或电流；
4）将主动驱动器通过腱拖动关节进而拖动可控负载的运动过程作为控制对象；该控制对象等效为单驱动器驱动的机器人关节运动控制过程，因此将控制问题有效地简化了；
5）将由灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量，将关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量，形成开环或闭环的控制。
参见附图5，所述基于动态模型的控制策略为：
1）建立构成拮抗式驱动的各个驱动器、传动装置、关节和外部负载的动态模型17；所述动态模型17包括构成拮抗式驱动的各个驱动器的电压或电流、关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及时间之间的函数关系；在缺少关节力和力矩传感器3或腱张力传感器5的情况下能够通过所述动态模型17对关节力和力矩、关节阻尼、关节刚度、腱张力进行估计；
2）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行开环控制时，将目标量和关节位置信息作为所述动态模型17的因变量，求解得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流（抑或是电压或电流随时间变化的函数）；
3）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行闭环控制时，将关节位置信息、构成拮抗式驱动的各个驱动器各自的电压或电流作为所述动态模型17的输入，所述动态模型17的输出即为关节力和力矩、关节阻尼、关节刚度、腱张力的估计值，将目标量对应的估计值作为反馈量，并将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元16；
4）当对目标关节位置、目标关节速度作为目标量进行闭环控制时，将关节位置信息和关节速度信息作为反馈量，将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元16；
5）经控制单元16进一步计算得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流。
所述控制单元16可以采用PID控制率或基于神经网络的控制率。
所述基于神经网络的控制策略为：
1）构建多层神经网络作为运动控制器；所述多层神经网络可采用模拟生物脑的基底核与小脑的脉冲神经网络；
2）将由灵巧手的一至多个关节的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量输入给所述多层神经网络；
3）将由灵巧手的一至多个关节的关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量输入给所述多层神经网络；
4）经过所述多层神经网络计算得到的输出即为各个驱动器分别应施加的电压或电流。
以上仅为本申请的可选实施例而已，并不用于限制本申请。对于本领域的技术人员来说，本申请可以有各种更改和变化。凡在本申请的精神和原则之内，所作的任何修改、等同替换、改进等，均应包含在本申请的权利要求范围之内。

Claims (10)

  1. 一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，包括：由腱传动的拮抗式驱动的灵巧手、传感器模块、传感器管理模块、驱动器控制模块、中央控制模块；
    所述由腱传动的拮抗式驱动的灵巧手配置为具有一至多个由腱传动并且采用拮抗式驱动的关节的灵巧手；
    所述传感器模块包括由多个关节角度传感器构成的关节角度传感器集合、由仿生皮肤中的多个触觉传感器构成的触觉传感器集合、由多个关节力和力矩传感器构成的关节力和力矩传感器集合、由多个腱张力传感器构成的腱张力传感器集合；
    所述关节角度传感器安装于灵巧手的各个关节处，测量各个关节的旋转角度，其输出信号经所述传感器管理模块处理得到关节位置信息；
    所述触觉传感器分布于仿生皮肤中，感知与物体的接触情况，其输出信号经所述传感器管理模块处理得到触觉信息；
    所述关节力和力矩传感器安装于灵巧手的各个指间关节、掌指关节、腕掌关节、腕部关节处，测量关节处的一至多维力或力矩，其输出信号经所述传感器管理模块处理得到关节力和力矩信息；
    所述腱张力传感器安装在腱上，测量腱的张力，其输出信号经所述传感器管理模块处理得到腱张力信息；
    所述传感器管理模块对传感器模块中的各个关节角度传感器、各个触觉传感器、各个关节力和力矩传感器、各个腱张力传感器施加恒定电源或以周期扫描的方式施加电源，并对这些传感器的输出信号进行放大、滤波、采样与转换，以及监测是否有传感器缺失或工作异常，并将处理过的输出信号与监测结果传给中央控制模块；
    所述传感器管理模块能够接收中央控制模块传来的控制指令，并根据控制指令调整工作方式；
    所述驱动器控制模块具有控制各个驱动器的电流环、电压环、速度环，自动对驱动器的电压或电流进行过载保护，监测各个驱动器是否缺失或工作异常，并将各个驱动器的电流、电压、速度和监测结果传给中央控制模块；
    所述驱动器控制模块通过所述传感器管理模块读取关节角度信息，构成关节限位直接控制环路；
    所述驱动器控制模块通过所述传感器管理模块读取关节力和力矩信息，构成关节力和力矩保护直接控制环路；
    所述驱动器控制模块通过所述传感器管理模块读取腱张力信息，构成腱张力保护直接控制环路；
    所述驱动器控制模块能够接收中央控制模块传来的控制指令，并根据控制指令调整各个驱动器的控制方式，即单独选择电流环、电压环、速度环或选择它们的任意组合对驱动器进行控制；
    所述中央控制模块接收操作目标，控制传感器管理模块并读取各个传感器的信息，经过多传感器信息综合与控制策略计算过程得到控制信号，将控制信号传给驱动器控制模块以控制各个驱动器，进而对灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及与物体的接触情况进行控制；
    所述中央控制模块搭载有传感器采样与分析策略、防止腱松弛的控制策略、防止腱过紧的控制策略、基于可控负载的控制策略、基于动态模型的控制策略以及基于神经网络的控制策略；
    所述基于可控负载的控制策略为，将对受控关节构成拮抗式驱动的各个驱动器分为主动驱动器和从动驱动器，通过调整从动驱动器的电压和/或电流使其处于跟随运动模式，使从动驱动器通过腱被关节拖动从而等效为可控负载，形成开环或闭环控制；
    所述基于动态模型的控制策略为，建立构成拮抗式驱动的一至多个驱动器、传动装置、关节和/或外部负载的动态模型，通过动态模型对一至多个状态变量进行估计，形成开环或闭环控制；
    所述基于神经网络的控制策略为，采用神经网络作为控制器，将一至多个传感信息或状态变量输入至所述神经网络，将所述神经网络的输出作为一至多个驱动器的控制量；
    所述中央控制模块能够通过配置指定对特定关节采用所述基于可控负载的控制策略或所述基于动态模型的控制策略或所述基于神经网络的控制策略；
    在灵巧手没有安装关节力和力矩传感器或腱张力传感器的情况下，所述中央控制模块默认采用所述基于动态模型的控制策略或所述基于可控负载的控制策略；
    在灵巧手安装了关节力和力矩传感器以及腱张力传感器的情况下，所述中央控制模块默认采用所述基于神经网络的控制策略；当关节力和力矩传感器或腱张力传感器部分或全部缺失或工作异常时，所述中央控制模块自动切换为采用所述基于动态模型的控制策略或所述基于可控负载的控制策略，以保证系统可靠；
    所述传感器采样与分析策略包括下面的一种或任几种：
    1）在需要感知和物体的接触情况，或需要感知物体的形状、纹理和质地，而不需要精确测量接触力的大小时，优先对仿生皮肤中的触觉传感器进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大触觉信息相对其它传感信息的权重；
    2）在需要精细操作并需要精确感知手部各个指节和关节与物体接触的相互作用力和力矩时，优先对关节力和力矩传感器进行采样，提高其采样频率，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重，同时能够配合仿生皮肤中的触觉传感器的信息进一步感知手部和物体的接触情况，以及感知物体的形状、纹理和质地；
    3）在做拉扯、扳扣以及拎提重物需要预设力道的动作时，优先对腱张力传感器进行采样，提高其采样频率，以估测各个手指或手部整体对物体施加的力，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重，同时能够配合关节力和力矩传感器的信息进一步对关节力或力矩进行精确测量；
    4）在仿生皮肤的部分或全部触觉传感器缺失或失效的情况下，优先对关节力和力矩传感器进行采样，并在多传感器信息综合与控制策略计算过程中放大关节力和力矩信息相对其它传感信息的权重；
    5）在部分或全部关节力和力矩传感器缺失或失效的情况下，优先对腱张力传感器进行采样，并在多传感器信息综合与控制策略计算过程中放大腱张力信息相对其它传感信息的权重。
  2. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述传感器管理模块对各个传感器施加电源的频率和对其输出信号的采样频率进行自动调节，其调节方式为：
    1）当某个传感器的输出信号幅度在一定时间范围内持续低于敏感阈值，或输出信号幅度随时间的变化率在一定时间范围内持续低于变化率阈值，则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第一预设频率范围；
    2）当某个传感器的输出信号幅度高于敏感阈值，或输出信号幅度随时间的变化率高于变化率阈值，则将对该传感器施加电源的频率和对其输出信号的采样频率调整至第二预设频率范围。
  3. 根据权利要求1或权利要求2所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述传感器管理模块在接收到中央控制模块的控制指令时，其工作方式包括下面的一种或任几种：
    1）针对控制指令指定的传感器以指定的频率施加电源；
    2）针对控制指令指定的传感器进行监测；
    3）以控制指令指定的倍数放大指定的传感器的输出信号；
    4）以控制指令指定的滤波方式对指定的传感器的输出信号进行滤波；
    5）优先对控制指令指定的传感器的输出信号进行采样与转换；
    6）以控制指令指定的采样频率对指定的传感器的输出信号进行采样与转换。
  4. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，当系统发生异常，使关节超出了其容许的运动范围时，所述关节限位直接控制环路进行快速响应，直接控制驱动器将关节限制在容许的运动范围内；所述中央控制模块具有禁止关节限位直接控制环路的控制指令，能够主动阻止关节限位直接控制环路工作。
  5. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，当系统发生异常，使关节力和力矩超出了其容许的范围时，所述关节力和力矩保护直接控制环路进行快速响应，直接控制驱动器将关节力和力矩限制在容许的范围内；所述中央控制模块具有禁止关节力和力矩保护直接控制环路的控制指令，能够主动阻止关节力和力矩保护直接控制环路工作。
  6. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，当系统发生异常，使腱张力超出了其容许的范围时，所述腱张力保护直接控制环路进行快速响应，直接控制驱动器将腱张力限制在容许的范围内；所述中央控制模块具有禁止腱张力保护直接控制环路的控制指令，能够主动阻止腱张力保护直接控制环路工作。
  7. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述防止腱松弛的控制策略为：
    1）设定腱张力的最低阈值；
    2）通过腱张力传感器判断腱张力是否低于最低阈值；
    3）若腱张力低于最低阈值，则使牵引该腱的驱动器进一步收紧腱，直至腱张力等于或略高于最低阈值；
    所述防止腱过紧的控制策略为：
    1）设定腱张力的最高阈值；
    2）通过腱张力传感器判断腱张力是否高于最高阈值；
    3）若腱张力高于最高阈值，则使牵引该腱的驱动器进一步释放腱，直至腱张力等于或略低于最高阈值。
  8. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述基于可控负载的控制策略为：
    1）将对受控关节构成拮抗式驱动的各个驱动器分为主动驱动器和从动驱动器；
    2）通过调整从动驱动器的电压或电流使其处于跟随运动模式，使从动驱动器通过腱被关节拖动从而等效为具有惯性和能够控制阻尼的可控负载；
    3）根据关节速度信息以及可控负载所需产生的阻尼大小，通过驱动器的模型计算得到从动驱动器应施加的电压或电流；
    4）将主动驱动器通过腱拖动关节进而拖动可控负载的运动过程作为控制对象；
    5）将由灵巧手的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量，将关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量，形成开环或闭环的控制。
  9. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述基于动态模型的控制策略为：
    1）建立构成拮抗式驱动的各个驱动器、传动装置、关节和外部负载的动态模型；所述动态模型包括构成拮抗式驱动的各个驱动器的电压或电流、关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力以及时间之间的函数关系；在缺少关节力和力矩传感器或腱张力传感器的情况下能够通过所述动态模型对关节力和力矩、关节阻尼、关节刚度、腱张力进行估计；
    2）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行开环控制时，将目标量和关节位置信息作为所述动态模型的因变量，求解得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流；
    3）当对由关节力和力矩、关节阻尼、关节刚度、腱张力构成的集合或其子集作为目标量进行闭环控制时，将关节位置信息、构成拮抗式驱动的各个驱动器各自的电压或电流作为所述动态模型的输入，所述动态模型的输出即为关节力和力矩、关节阻尼、关节刚度、腱张力的估计值，将目标量对应的估计值作为反馈量，并将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元；
    4）当对目标关节位置、目标关节速度作为目标量进行闭环控制时，将关节位置信息和关节速度信息作为反馈量，将反馈量与目标量做差得到偏差量，将偏差量输入给控制单元；
    5）经控制单元进一步计算得到构成拮抗式驱动的各个驱动器分别应施加的电压或电流。
  10. 根据权利要求1所述的一种基于多传感器与拮抗式驱动的灵巧手控制系统，其特征在于，所述基于神经网络的控制策略为：
    1）构建多层神经网络作为运动控制器；
    2）将由灵巧手的一至多个关节的关节位置、关节速度、关节力和力矩、关节阻尼、关节刚度、腱张力、与物体的接触情况构成的集合或其子集作为目标量输入给所述多层神经网络；
    3）将由灵巧手的一至多个关节的关节位置信息、关节速度信息、关节力和力矩信息、腱张力信息、触觉信息构成的集合或其子集作为反馈量输入给所述多层神经网络；
    4）经过所述多层神经网络计算得到的输出即为各个驱动器分别应施加的电压或电流。
PCT/CN2020/133437 2019-12-02 2020-12-02 一种基于多传感器与拮抗式驱动的灵巧手控制系统 WO2021110062A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/831,091 US20220288775A1 (en) 2019-12-02 2022-06-02 Dexterous hand control system based on multi-sensor and antagonistic drive

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201911213420.7A CN110842952B (zh) 2019-12-02 2019-12-02 一种基于多传感器与拮抗式驱动的灵巧手控制系统
CN201911213420.7 2019-12-02

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/831,091 Continuation US20220288775A1 (en) 2019-12-02 2022-06-02 Dexterous hand control system based on multi-sensor and antagonistic drive

Publications (1)

Publication Number Publication Date
WO2021110062A1 true WO2021110062A1 (zh) 2021-06-10

Family

ID=69607152

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/133437 WO2021110062A1 (zh) 2019-12-02 2020-12-02 一种基于多传感器与拮抗式驱动的灵巧手控制系统

Country Status (3)

Country Link
US (1) US20220288775A1 (zh)
CN (1) CN110842952B (zh)
WO (1) WO2021110062A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE2250586A1 (en) * 2022-05-16 2023-11-17 Bioservo Tech Ab A strengthening glove, a control system, and methods for operating an actuating means

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11787050B1 (en) * 2019-01-01 2023-10-17 Sanctuary Cognitive Systems Corporation Artificial intelligence-actuated robot
DE102019107848B4 (de) * 2019-03-27 2022-06-15 Franka Emika Gmbh Robotergreifer und Verfahren zum Betrieb eines Robotergreifers
CN110842952B (zh) * 2019-12-02 2020-12-29 深圳忆海原识科技有限公司 一种基于多传感器与拮抗式驱动的灵巧手控制系统
CN114536382B (zh) * 2022-04-26 2022-08-12 中国科学院自动化研究所 神经拟态灵巧手机器人
CN117621030A (zh) * 2022-08-09 2024-03-01 深圳忆海原识科技有限公司 机器人多级控制系统

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100152898A1 (en) * 2008-12-15 2010-06-17 Gm Global Technology Operations, Inc. Joint-space impedance control for tendon-driven manipulators
US20110071664A1 (en) * 2009-09-22 2011-03-24 Gm Global Technology Operations, Inc. Human grasp assist device and method of use
CN104191429A (zh) * 2014-07-28 2014-12-10 南京航空航天大学 一种腱驱动机械手位置和腱张力的混合控制方法及控制装置
CN104755041A (zh) * 2012-11-02 2015-07-01 直观外科手术操作公司 用于医疗器械的自对抗驱动装置
US20170129110A1 (en) * 2015-11-05 2017-05-11 Irobot Corporation Robotic fingers and end effectors including same
CN109591041A (zh) * 2017-10-02 2019-04-09 斯寇司株式会社 指状机构、机械手和机械手控制方法
CN110842952A (zh) * 2019-12-02 2020-02-28 深圳忆海原识科技有限公司 基于多传感器的灵巧手拮抗式控制系统

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8483880B2 (en) * 2009-07-22 2013-07-09 The Shadow Robot Company Limited Robotic hand
CN103426351B (zh) * 2013-07-11 2018-05-18 牛欣 可远程复现的心动脉应脉诊训练装置及方法

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20100152898A1 (en) * 2008-12-15 2010-06-17 Gm Global Technology Operations, Inc. Joint-space impedance control for tendon-driven manipulators
US20110071664A1 (en) * 2009-09-22 2011-03-24 Gm Global Technology Operations, Inc. Human grasp assist device and method of use
CN104755041A (zh) * 2012-11-02 2015-07-01 直观外科手术操作公司 用于医疗器械的自对抗驱动装置
CN104191429A (zh) * 2014-07-28 2014-12-10 南京航空航天大学 一种腱驱动机械手位置和腱张力的混合控制方法及控制装置
US20170129110A1 (en) * 2015-11-05 2017-05-11 Irobot Corporation Robotic fingers and end effectors including same
CN109591041A (zh) * 2017-10-02 2019-04-09 斯寇司株式会社 指状机构、机械手和机械手控制方法
CN110842952A (zh) * 2019-12-02 2020-02-28 深圳忆海原识科技有限公司 基于多传感器的灵巧手拮抗式控制系统

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
SE2250586A1 (en) * 2022-05-16 2023-11-17 Bioservo Tech Ab A strengthening glove, a control system, and methods for operating an actuating means
WO2023224536A1 (en) * 2022-05-16 2023-11-23 Bioservo Technologies Ab A strengthening glove, a control system, and methods for operating an actuating means

Also Published As

Publication number Publication date
US20220288775A1 (en) 2022-09-15
CN110842952A (zh) 2020-02-28
CN110842952B (zh) 2020-12-29

Similar Documents

Publication Publication Date Title
WO2021110062A1 (zh) 一种基于多传感器与拮抗式驱动的灵巧手控制系统
Carrozza et al. Experimental analysis of an innovative prosthetic hand with proprioceptive sensors
Wu et al. “Hold-and-manipulate” with a single hand being assisted by wearable extra fingers
Hosseini et al. A sEMG-driven soft exosuit based on twisted string actuators for elbow assistive applications
JP5492281B2 (ja) 手袋ベース把持補助デバイスの制御
US9498401B2 (en) Robotic system for simulating a wearable device and method of use
JP6055985B2 (ja) パワーアシストロボット
CN106584093A (zh) 工业机器人自我装配系统及方法
Vitiello et al. A sensorless torque control for Antagonistic Driven Compliant Joints
Yang et al. Arm rehabilitation robot impedance control and experimentation
CN113633521B (zh) 一种上肢外骨骼康复机器人控制系统及控制方法
US8483879B2 (en) Human-robot interactive system having a human stiffness estimation control algorithm
Mefoued et al. Knee joint movement assistance through robust control of an actuated orthosis
CN114536382B (zh) 神经拟态灵巧手机器人
Zhao et al. Development of a multi-DOF anthropomorphic prosthetic hand
Hosseini et al. Development of sEMG-driven assistive devices based on twisted string actuation
Peternel et al. After a decade of teleimpedance: A survey
CN1736521A (zh) 手腕及手指关节运动康复训练机器人
CN116852336A (zh) 一种面向双臂机器人的上身遥操作力反馈设备
Yu et al. Jacobian analysis for parallel mechanism using on human walking power assisting
JPH02256489A (ja) 多関節ハンド
Eusebi et al. The UB Hand II control system: design features and experimental results
Poulton et al. Progress of a modular prosthetic arm
JP2014057628A (ja) パワーアシストロボット
Ruiz et al. Exoskeleton-based robotic platform applied in biomechanical modelling of the human upper limb

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20895950

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 11/11/2022)

122 Ep: pct application non-entry in european phase

Ref document number: 20895950

Country of ref document: EP

Kind code of ref document: A1