CN-121989240-A - Three-stage mixed mapping action resolving method for underactuated dexterous hand

CN121989240ACN 121989240 ACN121989240 ACN 121989240ACN-121989240-A

Abstract

The invention provides a three-stage mixed mapping action resolving method for an underactuated dexterous hand, belongs to the technical field of robot control and dexterous hand control, and can partially solve the problems of inconsistent high degree of freedom of the hand and low degree of freedom dimension of the underactuated dexterous hand, unstable mapping and poor task adaptability in the prior art. The method comprises the steps of collecting hand motion data, conducting preprocessing such as filtering, smoothing and normalization, conducting dimension reduction processing on the preprocessed hand motion to obtain low-dimension control quantity, generating underactuated flexible hand joint motion based on the low-dimension control quantity, conducting feasibility processing by combining joint limit, underactuated coupling, collision and other constraints, identifying gesture types such as grasping, pinching, pointing and the like, conducting task priority correction according to the gestures to obtain target motion, converting the target motion into control instruction output, and achieving real-time stable following of the hand motion by the underactuated flexible hand.

Inventors

CHEN ZHOUYI
LIU RONGCHEN
YIN BAOHUA
LI JIAN
CHEN QINGYU
LU ZHUANG

Assignees

江苏金智人形机器人科技有限公司

Dates

Publication Date: 20260508
Application Date: 20260202

Claims (10)

1. The three-stage mixed mapping action resolving method for the underactuated dexterous hand is characterized by comprising the following steps of: S1, collecting hand motion data and preprocessing the hand motion data; S2, performing dimension reduction processing on the preprocessed hand motion data to obtain a low-dimension control quantity representing the hand motion; S3, generating joint actions of the underactuated dexterous hand according to the low-dimensional control quantity, and performing constraint processing on the joint actions so as to meet execution conditions of the underactuated dexterous hand; s4, recognizing gesture types corresponding to the hand motions, and correcting task priority of the joint motions meeting execution conditions according to the gesture types to obtain target motions of the underactuated dexterous hand; S5, converting the target action into a control instruction of the underactuated dexterous hand and outputting the control instruction to realize the following of the underactuated dexterous hand to the hand action.
2. The method for resolving three-stage hybrid mapping motion for an underactuated dexterous hand according to claim 1, wherein in step S1, the hand motion data is acquired by one or more of a data glove, an optical motion capturing device, a virtual reality controller and an inertial measurement unit glove, and the acquired hand motion data comprises at least two of the following data: Each joint angle data of a human hand, fingertip three-dimensional position data, palm three-dimensional posture data, joint angular velocity data and joint angular acceleration data.
3. The method for resolving a three-stage hybrid mapping motion for an underactuated dexterous hand according to claim 1, wherein in the step S1, the preprocessed output frequency of the human hand motion data is identical to or not lower than the output frequency of the control command, and the preprocessing includes at least two of the following processes: performing low pass filtering or Kalman filtering on the hand motion data to suppress noise; performing time synchronization and resampling on the hand motion data to unify sampling periods; Sliding window smoothing is carried out on the hand motion data so as to reduce high-frequency jitter; Scale normalization and bias correction are performed on the human hand motion data to eliminate different operator differences.
4. The method for resolving a three-stage hybrid mapping action for an underactuated dexterous hand according to claim 1, wherein in the step S2, the dimension reduction process comprises the steps of: establishing a low-dimensional basis space of the human hand action; Projecting the high-dimensional vector of the human hand action to the low-dimensional basis space to obtain the low-dimensional control quantity; the dimension of the low-dimensional control quantity is smaller than the vector dimension of the hand motion and is used as the core control input of the underactuated dexterous hand, so that the motion redundancy elimination and noise immunity enhancement of the underactuated dexterous hand are realized.
5. The method for resolving a three-stage hybrid mapping motion for an underactuated dexterous hand according to claim 4, wherein in the step S3, the joint motion is one or more of a joint angle, a joint angular velocity, or a joint angle trajectory, and generating the joint motion of the underactuated dexterous hand comprises the steps of: Constructing a mapping relation from a low-dimensional control space to the under-actuated dexterous hand joint space based on the low-dimensional control quantity to obtain the joint initial action of the under-actuated dexterous hand; Constraint processing is carried out on the initial joint action, so that the output action of the underactuated smart hand meets at least two constraints of joint limit constraint, underactuated coupling constraint and self-collision constraint of the underactuated smart hand; wherein the under-actuated coupling constraint is determined by a driving coupling relationship between a driver input and a plurality of joint outputs of the under-actuated dexterous hand.
6. The method of three-stage hybrid mapping action resolution for an underactuated dexterous hand according to claim 5, wherein in step S3, the constraint processing includes: taking a feasible region of the underactuated dexterous hand as a constraint set, adopting a projection operator to project the joint initial motion of the underactuated dexterous hand into the feasible region, or adopting constraint optimization solution to obtain the joint motion meeting the constraint; The constraint optimization is a quadratic optimization or equivalent lightweight optimization form so as to meet the real-time control calculation requirement and reduce the occurrence probability of infeasible solutions and jump solutions, and the feasible domain comprises the joint limit constraint, the underactuated coupling constraint and the self-collision constraint.
7. The method for resolving a three-stage hybrid mapping action for an underactuated dexterous hand according to claim 1, wherein in the step S4, the gesture class recognition comprises the steps of: Constructing a gesture feature vector based on the human hand motion data; inputting the gesture feature vector into a classifier to obtain a gesture class; Taking the recognition result of the gesture class as a triggering condition for task priority correction; the gesture feature vector comprises at least two of joint angle features, joint angular velocity features, fingertip relative distance features and palm gesture features, the gesture category comprises one or more of grasping, pinching, pointing, finger dividing and swinging, the action following item is used for restraining deviation between the target action and the joint action output in the step S3, and the task item is further adjusted according to the size, shape or gesture of the target object so as to realize task self-adaptive control.
8. The method for resolving a three-stage hybrid mapping action for an underactuated dexterous hand according to claim 1, wherein in the step S4, the task priority correction includes: On the premise of meeting the constraint processing conditions, taking the joint action output in the step S3 as an initial value, and constructing an objective function comprising an action following item and a task item; obtaining a target action of the underactuated dexterous hand by optimizing and solving the target function; the task item is used for describing key action constraints matched with gesture categories, wherein the key action constraints comprise one or more of a grasping envelope constraint, a fingertip clamping constraint, a pointing direction constraint, an inter-finger unfolding constraint and a palm gesture constraint.
9. The method for resolving a three-stage hybrid mapping action for an underactuated dexterous hand according to claim 8, wherein in the step S4, the task priority correction adopts different task item setting modes for different gesture categories, and the method comprises: When the gesture type is gripping, the task item is used for increasing the contact envelope coverage degree of the underactuated dexterous hand and the target object, when the gesture type is pinching, the task item is used for restraining the relative pose of the thumb and the index finger to improve the clamping precision, when the gesture type is pointing, the task item is used for restraining the direction of pointing the finger to be consistent with the palm pose, when the gesture type is pointing, the task item is used for increasing the relative angle or the relative distance between the fingers, and when the gesture type is swinging, the task item is used for dynamically compensating the palm pose to improve the naturalness and the continuity of actions.
10. The underactuated dexterous hand oriented three-phase hybrid mapping action resolution method of any of claims 1 to 9, wherein in step S4 the task priority modification further comprises an approximate envelope strategy comprising the steps of: converting the control target of the underactuated dexterous hand from joint angle matching to key contact point position meeting or key contact point direction meeting; And carrying out flexible constraint or relaxation treatment on the non-critical degree of freedom of the underactuated dexterous hand, ensuring the action stability of the underactuated dexterous hand, and improving the completion rate of the grabbing or pinching task of the underactuated dexterous hand.

Description

Three-stage mixed mapping action resolving method for underactuated dexterous hand Technical Field The invention belongs to the technical field of robot control and dexterous hand control, and particularly relates to a three-stage mixed mapping action resolving method for underactuated dexterous hand. Background Along with the development of humanoid robots, remote control and man-machine interaction technologies, smart hands are used as key end effectors for performing grabbing, operating and fine working by robots, and are widely applied to scenes such as industrial collaboration, service robots, medical assistance and dangerous environment working. In order to improve the operation naturalness and control precision of the dexterous hand, the hand movement mapping control based on motion capture becomes an important research direction. In the existing system, a human hand has 20-26 degrees of freedom, can finish complex actions such as grasping, pinching, pointing and finger dividing, and most underactuated smart hands have 4-10 degrees of freedom, and constraint such as joint coupling, limiting and collision exists, so that high-dimensional actions of the human hand are difficult to map to executable spaces of the mechanical hand directly, and action distortion or unstable control easily occurs. The current common mapping method comprises joint angle direct mapping, gesture template matching, high-dimensional mapping based on optimization and the like. The joint angle direct mapping is sensitive to noise, shake or jump is easy to generate, unreachable gestures can be output, gesture template matching actions are discrete and poor in continuity, different object and task requirements are difficult to adapt, high-dimensional optimization mapping calculation amount is large, instantaneity is insufficient, infeasibility solution or difficulty in convergence is easy to occur under underactuated conditions, and therefore, the three-stage mixed mapping action resolving method for underactuated smart hands is provided. Disclosure of Invention The invention aims to at least solve one of the technical problems in the prior art and provides a three-stage mixed mapping action resolving method for underactuated dexterous hands. The invention provides a three-stage mixed mapping action resolving method for an underactuated smart hand, which comprises the following steps of: S1, collecting hand motion data and preprocessing the hand motion data; S2, performing dimension reduction processing on the preprocessed hand motion data to obtain the low-dimension control quantity representing the hand motion; S3, generating joint actions of the underactuated dexterous hand according to the low-dimensional control quantity, and performing constraint processing on the joint actions so as to meet execution conditions of the underactuated dexterous hand; s4, recognizing gesture types corresponding to hand actions, and correcting task priority of the joint actions meeting execution conditions according to the gesture types to obtain target actions of the underactuated dexterous hand; S5, converting the target action into a control instruction of the underactuated dexterous hand and outputting the control instruction to realize the following of the underactuated dexterous hand to the hand action. Further, in the step S1, the hand motion data is acquired by one or more of a data glove, an optical motion capturing device, a virtual reality controller and an inertial measurement unit glove, and the acquired hand motion data includes at least two of the following data: Each joint angle data of a human hand, fingertip three-dimensional position data, palm three-dimensional posture data, joint angular velocity data and joint angular acceleration data. Specifically, in the step S1, the preprocessed output frequency of the hand motion data is identical to or not lower than the output frequency of the control command, and the preprocessing includes at least two of the following processes: performing low pass filtering or Kalman filtering on the hand motion data to suppress noise; performing time synchronization and resampling on the hand motion data to unify sampling periods; Sliding window smoothing is carried out on the hand motion data so as to reduce high-frequency jitter; Scale normalization and bias correction are performed on the human hand motion data to eliminate different operator differences. Specifically, in the step S2, the dimension reduction process includes the following steps: establishing a low-dimensional basis space of the human hand action; Projecting the high-dimensional vector of the human hand action to the low-dimensional basis space to obtain the low-dimensional control quantity; the dimension of the low-dimensional control quantity is smaller than the vector dimension of the hand motion and is used as the core control input of the underactuated dexterous hand, so that the motion redundancy elimination and noise immu