Classifications

• • B—PERFORMING OPERATIONS; TRANSPORTING
  • B25—HAND TOOLS; PORTABLE POWER-DRIVEN TOOLS; MANIPULATORS
  • B25J—MANIPULATORS; CHAMBERS PROVIDED WITH MANIPULATION DEVICES
  • B25J9/00—Program-controlled manipulators
  • B25J9/16—Program controls
  • B25J9/1694—Program controls characterised by use of sensors other than normal servo-feedback from position, speed or acceleration sensors, perception control, multi-sensor controlled systems, sensor fusion
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B2219/00—Program-control systems
  • G05B2219/30—Nc systems
  • G05B2219/39—Robotics, robotics to robotics hand
  • G05B2219/39529—Force, torque sensor in wrist, end effector
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B2219/00—Program-control systems
  • G05B2219/30—Nc systems
  • G05B2219/40—Robotics, robotics mapping to robotics vision
  • G05B2219/40541—Identification of contact formation, state from several force measurements
• • G—PHYSICS
  • G05—CONTROLLING; REGULATING
  • G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
  • G05B2219/00—Program-control systems
  • G05B2219/30—Nc systems
  • G05B2219/40—Robotics, robotics mapping to robotics vision
  • G05B2219/40583—Detect relative position or orientation between gripper and currently handled object

Definitions

• the present invention relates to a method (and corresponding entity, computer program, system, medium) for determining an in-hand pose of an object held by a robot.
• estimating the pose of a grasped object is an important requirement and task. For instance, in a peg-in-hole insertion task, the robot should know the pose of the peg in its hand precisely in order to know the relative position and pose between the peg and the hole; once this is known, then the robot can insert the peg into the hole easily using, for instance, forward/inverse kinematics and impedance control.
• the robot should know the pose of the peg in its hand precisely in order to know the relative position and pose between the peg and the hole; once this is known, then the robot can insert the peg into the hole easily using, for instance, forward/inverse kinematics and impedance control.
• uncertainty by noisy measurement in grasped pose estimation causes failures in subsequent manipulation tasks.
• prior art techniques often rely on computationally complex models to perform such a task, or on a number of information about the robot and/or the environment that are not always available or easy to obtain, thus decreasing the reliability of and repeatability in performing the task, and increasing the setup cost.
• a method for determining an in-hand pose of an object held by a robot comprising the steps of moving, in an environment, a hand of the robot holding an object; estimating, at a given time instant, an in-hand pose with which the robot is holding the object based on earlier interaction information, the earlier interaction information including position information indicating a hand position at an earlier time instant, and interaction status information indicating whether the robot has interacted with an environment reference element at the earlier time instant, the earlier time instant being a time instant preceding in time the given time instant.
• the position information indicates a plurality of hand positions at respective earlier time instants
• the status interaction information indicates whether the robot, in correspondence of said respective earlier time instants, has interacted with the environment reference element.
• the earlier interaction information includes interaction status information indicating that the robot has interacted with the environment reference element at least at one earlier time instant.
• the earlier interaction information comprises interaction status information indicating that the robot has interacted with the environment reference element at least at two, preferably at least three, earlier time instants, and the position information indicates the hand position at the corresponding at least two, and respectively three, interaction time instants.
• estimating the in-hand pose comprises determining a posterior probability of the in-hand pose based on the earlier interaction information.
• estimating the in-hand pose comprises calculating the in-hand pose based on wherein x represents the in-hand pose, c t the pose of the hand, and y t indicates whether the robot has interacted with the environment reference element.
• the posterior probability of the in-hand pose is determined using a particle filter method.
• the posterior probability of the in-hand pose is determined on the basis of:
• the particles of the particle filter method are updated based on an interaction status information for a given particle.
• the method of the first aspect of the invention includes: estimating, at a further earlier time instant preceding in time the earlier time instant, the in hand pose on the basis of information that is different from information depending on interaction with the environment.
• the estimating at the further earlier time instant includes estimating the in-hand pose on the basis of predetermined hand position information given in advance, and/or sensor information indicating an interaction between the hand and the object.
• estimating the in-hand pose includes estimating the in-hand pose on the basis of a model of the object, wherein the model of the object is preferably a geometrical model.
• the interaction status information is determined on the basis of sensor modality information.
• the interaction status information is based on the output of one or more amongst a force-torque sensor, a tactile sensor, an image sensor, a 3D sensor.
• an entity (200) for determining an in-hand pose of an object held by a robot comprising: a movement controller (210) for causing the hand of the robot to move in an environment, while the hand is holding an object; a processor (220) configured to estimate, at a given time instant, an in-hand pose with which the robot is holding the object based on earlier interaction information, the earlier interaction information including position information indicating a hand position at an earlier time instant, and interaction status information indicating whether the robot has interacted with the environment reference element at the earlier time instant, the earlier time instant being a time instant preceding in time the given time instant.
• the position information indicates a plurality of hand positions at respective earlier time instants
• the status interaction information indicates whether the robot, in correspondence of said respective earlier time instants, has interacted with the environment reference element.
• the earlier interaction information includes interaction status information indicating that the robot has interacted with the environment reference element at least at one earlier time instant.
• the earlier interaction information comprises interaction status information indicating that the robot has interacted with the environment reference element at least at two, preferably at least three, earlier time instants, and the position information indicates the hand position at the corresponding at least two, and respectively three, interaction time instants.
• the processor (220) is configured to estimate the in-hand pose by determining a posterior probability of the in-hand pose based on the earlier interaction information.
• the processor (220) is configured to estimate the in-hand pose by calculating the in-hand pose based on wherein x represents the in-hand pose, c t the pose of the hand, and y t indicates whether the robot has interacted with the environment reference element.
• the posterior probability of the in-hand pose is determined using a particle filter method.
• the posterior probability of the in-hand pose is determined on the basis of:
• the particles of the particle filter method are updated based on an interaction status information for a given particle.
• the processor is further configured to estimate, at a further earlier time instant preceding in time the earlier time instant, the in-hand pose on the basis of information that is different from information referring to interaction with the environment.
• the processor is configured to estimate at the further earlier time instant by estimating the in-hand pose on the basis of predetermined hand position information given in advance, and/or sensor information indicating an interaction between the hand and the object.
• the processor is configured to estimate the in-hand pose by estimating the in-hand pose on the basis of a model of the object, wherein the model of the object is preferably a geometrical model.
• the interaction status information is determined on the basis of sensor modality information.
• the interaction status information is based on the output of one or more amongst a force-torque sensor, a tactile sensor, an image sensor, a 3D sensor.
• the entity (200) of the second aspect of the invention is comprised by a robot.
• a computer program comprising instructions configured to perform, when said instructions are executed on a computer, any of the steps described in the above first embodiment and/or its optional aspects.
• a medium comprising instructions configured to execute, when said instructions are executed on a computer, any of the steps described in the above first embodiment and/or its optional aspects.
• Figure 1 is a flow chart illustrating a method according to an embodiment of the present invention
• Figure 2 is a flow chart illustrating a method according to a further embodiment of the present invention.
• Figure 3 shows a block diagram of a device according to one embodiment of the present invention
• Figure 4 is a schematic representation of a robot placed in an environment and holding an object, according to one example of the invention.
• Figure 5 shows a block diagram of a computer suitable for executing instructions according to one embodiment of the present invention.
• the invention is based, amongst others, on the inventors' recognition that an in-hand pose can be better estimated by considering possible interaction (s) between the robot, in particular the grasped object and/or the grasping hand, and the environment in which the robot is placed (or is in operation) or capable of moving.
• the in-hand pose indicates a position of an object held by the hand of a robot relatively to the hand (or more in general relatively to the robot), and may be expressed for instance by way of a 6-dimensional vector (this being a non limiting example, see also further below) comprising translation and rotation values, which may include values relative to the hand or robot, or values relative to a reference system being outside/independent of the hand/robot; thus, the in-hand pose provides an indication of how the object is placed relatively to the hand or robot.
• the in-hand position indicates a relative position (or relative placement) between a reference point of the object and a reference point of the robot (or of the robot's hand, or of another robot's part, as long as it is a reference point); preferably, the reference point indicates also an orientation of the object.
• robot it is intended any device capable of carrying out a series of actions, including movements (of the robot and/or of its extensions) and/or grasping object(s).
• a robot thus includes an industrial robot, a service robot, an interactive robot, a manipulator, etc.
• hand it is herein referred to a component of the robot capable of grasping and/or holding an object, preferably while the hand and/or robot is moving.
• the hand thus includes any device capable of grasping and/or holding an object, and thus includes fingers, an end-effector (in the sense of an extension of the robot capable of grasping and/or holding), etc.
• the term hand is not restricted to a particular shape or construction, like for instance an anthropomorphic shape/construction.
• the hand of the robot is moved while holding an object in an environment.
• the hand of the robot is moved along a trajectory; the movement or trajectory can be determined in a way that is for example random, according to predetermined rule(s), according to a predetermined pattern (s), programmed, under control of an operation, or any combination of these.
• environment it is intended the surroundings of the robot, like for instance: a table or surface (or part(s) thereof), on which the robot is installed and can operate; a room (or part(s) thereof), a factory (or part(s) thereof) in which the robot can operate, etc., as well as any combination of these.
• the estimation may be based on earlier interaction information, wherein the term earlier indicates that the interaction information refers to a point in time which precedes (in time) the given time instant at which the estimation is performed.
• the earlier in time also indicates that the interaction information represents data or evidence that has been obtained and it thus available at the time instant at which the estimation is made.
• the given time instant is a point in time at which the estimation is performed, preferably when taking into account tolerances implied in the determination of a point in time (i.e. the term is not restricted to an exact point in time, and may include a time interval for performing such estimation; similarly, the measurement or data are obtained at a time instant that may include a time interval as necessary for making measurements or acquisition).
• the above introduced earlier interaction information include position information and interaction status information .
• the position information indicates a hand position at an earlier time instant, wherein the earlier time instant is a time instant preceding in time the given time instant (at which the estimation is performed); in one example later illustrated, the given time instant is represented by T i and the earlier time instant by T i-1 .
• the position information therefore represents the position of the robot's hand at a point in time preceding the timing at which the estimation is performed, and may be obtained for instance by any suitable sensor as also below illustrated.
• the robot and the environment reference element are characterized by a relative (spatial) separation, which is preferably known or that can be calculated on the basis of reference point(s). This separation may be expressed in terms of (also relative) positions of the two elements, relative distance, relative angle, etc. or any combination of these.
• the environment reference element represents one element included in the environment, and which is preferably predetermined (hence, the term reference, which can be interchanged with predetermined); reference or predetermined element implies that the separation (in space) between the element and the robot (or parts thereof) is known, or can be calculated on the basis of predetermined point(s) of the robot and the environment.
• the element (and correspondingly the relative position/distance/angle between the element and the robot) can be dynamically changed, as long as the spatial separation between the robot and that element is known (or can be calculated).
• the robot may be placed on a table, with the surface (or a portion thereof) of the table representing an environment reference element; the relative height of one predetermined point of the robot from the table's surface is an example of a predetermined or reference separation between the reference element and the robot.
• the element may be represented by a wall, by a line or segment in the environment, by a specific point in the environment, or any combination thereof.
• the robot itself or a known portion or surface of the robot may be regarded as the reference element, and the relative separation may be the separation between the refence part of the robot (e.g. a part of the robot's basement, or a joint, etc.) and another reference point of the robot.
• the interaction status information indicates whether there has been any robot-environment interaction.
• robot-environment interaction can be either (i) a direct interaction, when for instance the robot (of one of its components including the hand) interacts with the environment or (ii) an indirect interaction, when for instance the object held by the hand interacts with the environment, or any of their combination.
• the object can be considered as a part of the robot for the purpose of determining the interaction.
• indirect touch one may consider the case wherein the object is a pen being held while pointing downwards, and the robot is placed on a table (an example of the environment); if the hand is moved towards the table, it is likely that the object interacts with the table (e.g. by touching the table) before the robot's hand interacts with the table; in this case, it can be determined that the object - and consequently the robot in an indirect manner - has interacted with the table.
• the hand of the robot or another component of the robot like e.g. an arm
• the interaction is represented by a touch state.
• a touch as herein used also includes the case wherein the robot/hand and/or object graze the environment, in the sense that the robot/hand and/or object is about to touch the environment but just misses touching the environment.
• a touch includes also the case where the robot goes very close to touching the environment, e.g. close to touching the environment within a given threshold.
• the interaction includes any action that physically couples the robot with the environment as a consequence of the movement of the grasped object. Suitable sensors can be used as also later illustrated.
• the method of Figure 1 can also be explained in other illustrative words: at time T i-1 , it is determined whether or not the robot interacts, e.g. touches in the above sense, the environment; at the same time instant T i-1 , the hand's position is also measured. Then, at time T i , the in-hand pose is estimated based on (i) the measurement at time T i-1 of the hand position and (ii) the information as to whether at that time instant T i-1 there was a robot-environment interaction.
• the position information indicates a plurality of hand positions at respective earlier time instants
• the status interaction information indicates whether the robot, in correspondence of the respective earlier time instants, has interacted with the environment reference element.
• the earlier interaction information includes interaction status information indicating that the robot has interacted with the environment reference element at least at one earlier time instant, which may be indicated as the interaction time instant.
• the position information thus indicates the position of the hand when the robot-environment interaction occurred.
• the earlier interaction information includes interaction status information indicating that the robot has interacted with the environment reference element at least at two, preferably at least at three, earlier time instants; the position information indicates the hand positions at the respective at least two, and preferably respectively three, interaction time instants.
• the in-hand pose can thus be estimated on the basis of information obtained in correspondence of at least two, preferably three, time instants at which the robot- environment interaction occurred.
• the in-hand pose can be estimated after a robot-environment interaction has occurred at least two, or correspondingly at least the three times, before the instant at which the estimation is computed.
• the in-hand pose estimate can be refined after each interaction.
• estimating the in-hand pose comprises determining a posterior probability of the in-hand pose based on the earlier interaction information. Examples will be provided also below, e.g. with reference to Baysian models.
• estimating the in-hand pose comprises calculating the in-hand pose based on wherein x represents (the variable indicating) the in hand pose, c t the (variable indicating) pose of the hand, and y t (the variable that) indicates whether the robot has interacted with the environment reference element. Examples and further illustrations will be given below.
• the posterior probability of the in-hand pose is determined using a particle filter method. Also here, examples and further illustrations will be given further below.
• the posterior probability of the in-hand pose is determined on the basis of:
• the particles of the filter method are updated based on an interaction status information for a given particle.
• the particles are updated on the basis of those particles for which an interaction (e.g. a touch) has occurred.
• an interaction e.g. a touch
• the updating of the particles is also called sometimes re-sampling.
• the further earlier time instant includes an initial time instant (e.g. when the method for estimating is started or about to be started) at which there are no measurements available such that predetermined information are used to initialize the method, i.e. predetermined information (that are not necessarily measurements but may be previous measurement and/or information obtained by other method(s)) are used to start the estimation process at the next time instant T i .
• the in-hand pose on the basis of predetermined hand position information given in advance (e.g. known), and/or sensor information indicating an interaction between the hand and the object.
• predetermined hand position information e.g. known
• sensor information indicating an interaction between the hand and the object.
• Different sensors are suitable for providing an indication of interaction between the hand and the object, as also further later illustrated.
• estimating the in-hand pose includes estimating the in-hand pose on the basis of a model (preferably, a geometrical model) of the object. Having in fact knowledge of the model, it is possible to determine more accurately the placement in space of the object while being held by the hand - it is then possible to perform tasks like for example the peg-in-hole task more accurately (noting that performing the task is optional, what matters being the in-hand pose estimation).
• a CAD model is an example of a model for the object.
• a model of the object is not necessarily required to obtain the in-hand pose, as in fact the object (or a class of objects) can be often approximated or represented with one single information, preferably known in advance or which can be assumed, estimated, or measured (or a combination thereof).
• the in-hand pose can be defined relatively to the sphere's centre; in the case of an elongated object (i.e. an object which predominantly extends over one dimension, like the length, making the other directions negligible, e.g.
• the in-hand pose can be defined relatively to one of the distal ends of the elongated object, or relatively to the centre or another point of the elongated object, etc.
• it is not strictly necessary to represent the in-hand pose with a 6 dimensional vector as for instance it may be enough to specify a distance between a reference point of the object and the reference point of the hand, or an angle between a reference dimension (e.g. longitudinal extension) of the object and a reference axis of the hand/robot.
• the information relating to the model does not need to be predetermined; for instance, image recognition could be performed on an object (as grasped, or (about) to be grasped by the hand), so that the object can be classified, recognized or estimated, such that the in-hand pose can be determined based on the classification/recognition/estimation of the object.
• image recognition could be performed on an object (as grasped, or (about) to be grasped by the hand), so that the object can be classified, recognized or estimated, such that the in-hand pose can be determined based on the classification/recognition/estimation of the object.
• above reference has been made to an analytical geometrical shape as a model, and in general to any model that, e.g. by taking the geometry into account, allows estimating the object's position in space.
• a model representing those modalities may be used, like for instance a model representing electric or thermal conductivity.
• the model may include its stiffness, elasticity, viscosity, etc. or any
• the interaction status information is determined on the basis of sensor modality information.
• the sensor modality information represents information provided by a sensor suitable for sensing sensor modality, i.e. a sensor suitable to detect one aspect of a stimulus or what is sensed/detectable after a stimulus.
• sensors are presented by a proximity sensor, an infrared light sensor, an electric contact sensor, etc. or a combination thereof.
• the interaction status information is based on the output of one or more amongst a force-torque sensor, a tactile sensor (including are sensors), an image sensor, a 3-D sensor, etc. or a combination thereof.
• a force-torque sensor output can be used to determine whether the object and/or robot has touched the environment.
• Image recognition on an image taken while the object is moved in the environment
• image recognition may be used to determine where the object is located of where its bounds are (so as to determine whether this results in an interaction with the environment), etc. (these methods may be combined). Similar considerations apply to other sensors.
• An initialisation is performed at step S110, which includes providing an initial estimation at time T 0 for the in-hand pose without relying on any measurement data; the initial estimation can be provided for example on the basis of a predetermined hand position given in advance, a predetermined (e.g. a priori) probability distribution for the in-hand pose like for instance a flat probability distribution, sensor information indicating an interaction between the hand and the object (as for instance given by a tactile sensor, an array sensor, etc.).
• the robot is controlled such that the hand holding the object is moved along a trajectory as also above explained.
• the initial estimation and the start of movement may be done at the same time or at different time instants.
• step S150 it is determined whether a robot-environment interaction occurs and respective interaction status information at time T1 is generated; for instance, "1" indicates that an interaction occurred, while "0" indicates that an interaction has not occurred (other representations are possible). Further, the position of the hand is also measured so that position information at time T 1 is obtained. In this way, it is possible obtaining earlier interaction information at time T 1 , which includes information on the hand position and on the robot-environment interaction at time T 1 .
• the robot-environment interaction and hand position are measured and refer to the same time instant T 1 (which, as said, especially when taking tolerances into account may be regarded as a measurement and/or detection period); the earlier time information are preferably prepared at the same time, but not necessarily as in fact they may be generated (shortly) after obtaining the measurements.
• an estimation of the in-hand pose is performed, by using the earlier interaction information obtained at time T 1 . Thanks to the contribution of the measurement data, the in-hand pose can be estimated with a certain accuracy. In case the interaction status information indicates that a robot- environment interaction occurred once, e.g. that the robot touched (or almost touched) the environment reference element, the in-hand pose can be estimated even far more accurately .
• step S150 at time instant T 2 so as to obtain earlier interaction status information at time instant T 2 ; such earlier interaction status information may include also the measurement at time instant T 1 , i.e. both of the data available at T 1 and T 2 may be included therein, or only the most recent data of time instant T 2 .
• the method can then proceed to step S160 wherein an estimation is performed at time instant T 3 by taking into account the newly available earlier interaction status information.
• the process can be further iterated at subsequent time instants T 3 , T 4 , ...
• the above method(s) can be executed by a computer, by processor, by a combination of processor(s) and/or processing unit(s), by a single device, by a plurality of interconnected devices (e.g. in a cloud), or a combination thereof.
• the entity 200 can be realised by any combination of software and/or hardware, and can be concentrated into one single device or distributed across multiple devices preferably interconnected (over any suitable communication network (s)) with each other.
• the entity 200 comprises a movement controller 210 and a processor 220.
• the hand movement controller 210 controls the hand of the robot so that it is moved in an environment where the robot is placed or is in operation, while the hand is holding the object. Thus, the movement controller causes the movement of the robot's hand.
• the movement may be according to a continuous trajectory or a set of different trajectories each representing a section (or stretch) of movement.
• the processor 220 is configured to estimate, at a given time instant T i , an in-hand pose with which the robot is holding the object based on earlier interaction information referring to an earlier time instant T i-1 .
• the earlier interaction information includes position information and interaction status information.
• the position information indicates a position of the hand at time instant T i-1 .
• the interaction status information indicates whether the robot has interacted with the environment reference element at instant T i-1 .
• the entity of the present embodiment is further configured to perform any of (and any combination of) the operations described above or in the following, such that repetitions are omitted.
• the entity 200 may be included within the robot, or placed remotely from the robot while being in communication with the same. In another example, the entity 200 may be distributed, for example partly placed within the robot and partly placed remotely, or realized in a distributed manner all remotely from the robot.
• FIG 4 shows an illustrative example of a robot configuration to which the present invention may be applied.
• robot R includes a hand M, including two fingers R1 and R2 suitable for grasping and holding an object 0.
• the object has an elongated shape, such that it can be approximated to a stick having length L, while neglecting the other dimensions.
• Other objects and shapes are possible.
• Robot R is placed (or is in operation) in an environment. More in particular, in the example, robot R is mounted on a table T (T may in general represent a surface, e.g. of the table, pavement, etc.).
• the environment may of course include other elements, like for instance walls, pieces of furniture, industrial tools, etc., which are not represented for simplicity.
• the table T represents a reference element of the environment, in the sense that it is known a relative separation between the robot and the environment reference element T.
• a relative separation between the robot R and the reference environment element T can be represented by a height H of the robot R from the table T. More in particular, the height H may be known with reference to a point PI of the robot R (when the robot R finds itself in a certain position or configuration) and a point P2 of the table T; the geometry and the kinematics will then allow determining the height H' in any other position or configuration that the robot and its components may assume while in operation, i.e. while the robot is moving. Once the object 0 is grasped, the robot is controlled to move the object in the environment.
• the robot-environment interaction as well as the corresponding hand position are detected or measured at time instant T i-1 ; the collected measurements or data are then used at time instant Ti to estimate the in-hand pose.
• the robot interacts with (e.g. touches) the environment, and in particular the environment reference element, the estimation of the in-hand pose can be highly increased. This may be explained as follows: When the robot R touches the table T, the relative separation between the table and the robot, which is known or that can be calculated (see above, based e.g. on movement/kinematic), represents an additional piece of available information that can be used to improve the in hand pose estimation.
• the in-hand pose estimation will then indicate a relative placement of the object relatively to the robot, in the sense of being relative to any part of the robot.
• the in-hand pose may be relative to the hand of the robot, and may for instance be indicated by a 6-dimensional vector providing translation and rotation of the object 0 relatively to the hand; in another example, the in-hand pose for the object 0 illustrated in figure 4 may be given by an information indicating whether the object 0 points upwards or the downwards relatively to the hand, i.e. it does not need be necessarily represented by a 6 dimensional vector.
• Figure 5 illustrates a block diagram exemplifying a computer (500) capable of running the aforesaid program.
• the computer (500) comprises a memory (530) for storing the program instructions and/or the data necessary for its execution, a processor (520) for carrying out the instructions themselves and an input/output interface (510).
• the instructions may be those or executing any of the steps, or any combination of the steps, herein illustrated.
• a medium for supporting a computer program configured to perform, when the program is run on a computer, one or a combination of the steps according to the method described above, e.g. with reference to the first embodiment.
• Examples of a medium are a static and/or dynamic memory, a fixed disk or any other medium such as a CD, DVD or Blue Ray.
• the medium also comprises a means capable of supporting a signal representing the instructions, including a means of cable transmission (ethernet, optical, etc.) or wireless transmission (cellular, satellite, digital terrestrial, etc.).
• a system including a robot and an entity as illustrated for instance with reference to the embodiment depicted in Figure 3.
• the entity may be placed remotely from the robot, or partially within or in proximity of the robot and partially outside the robot.
• the entity is integrated with or within the robot to form a system for estimating the in-hand pose, or is placed in proximity of (e.g. attached to) the robot.
• an in-hand pose estimation method is formulated by a particle filter framework to update estimate gradually by exploiting interactions between the grasped object and the environment.
• the robot makes a (at least one) contact between the grasped object and object(s) in the environment (above also called element(s) of the environment) such as a table to update the estimate from robot encoder information when it collides to the environment while knowing the table position in robot coordinates.
• the robot encoder information represents an example of the hand position information.
• x and c t are 6-dimensional vectors including translation (at Euclidean space coordinate Î R 3 ) and rotation (at Euler angle R 3 ), and y t is a binary value (Î ⁇ 0,1 ⁇ ) indicating contact or non-contact.
• the estimate in this example, can be updated using the particle filter, e.g. as follows.
• ⁇ y t , c t ⁇ T t 1) ⁇
• the posterior probability can be given by The above indicates that the posterior probability of the relative pose after the T-th interaction can be calculated by that after the T - 1-th interaction and the likelihood at T-the interaction with environment.
• the posterior probability cannot be computed in closed form, because the marginalization in the right-hand side of formula (2) cannot be computed in analytical way.
• the particle filtering method is used, one of the sampling-based approximations.
• the family of sampling-based methods gives guarantee of asymptotic convergence to the real posterior by increasing the number of particles, compared to variational methods or Kalman filtering.
• the objective is to achieve a highly precise estimation of the relative pose; accordingly, a sampling-based method has been chosen to obtain higher accuracy, rather than optimization- based methods which are advantageous in fast computation.
• marginalization in the right hand side of the formula (2) can be approximated, using particles of pose estimates sampled from the posterior probability at T-1 times, as follows:
• threshold is a small enough value, say about 0.5[mm]. This enables approximated posterior computations using particles; the next sampling of particles is done based on the posterior probability after updating .
• the likelihood for updated posterior probability can be computed as below, and the new particle is sampled in proportional probability based on the likelihood :
• each unit may be substituted by respective controlling means, memory means, processing means, sensing means, etc., respectively.
• These units can be implemented as distinct/self-contained units/entities or as distributed units/entities (i.e. implemented through a number of components connected to one another, whether physically near or remote); these, be they concentrated or distributed, can further be implemented through hardware, software or a combination thereof.

Landscapes

• Engineering & Computer Science (AREA)
• Robotics (AREA)
• Mechanical Engineering (AREA)
• Manipulator (AREA)

Priority Applications (2)

Applications Claiming Priority (1)

Publications (1)

Family

ID=68072905

Family Applications (1)

Country Status (2)

Cited By (1)

Families Citing this family (3)

Citations (1)

Family Cites Families (1)

• 2019
  • 2019-09-13 WO PCT/IB2019/057734 patent/WO2021048598A1/en not_active Ceased
  • 2019-09-13 JP JP2022514753A patent/JP2022546733A/ja active Pending

Patent Citations (1)

Non-Patent Citations (6)

Also Published As

Similar Documents

Legal Events