Search

CN-121989238-A - Motion control method, boxing control method and boxing control system for humanoid robot

CN121989238ACN 121989238 ACN121989238 ACN 121989238ACN-121989238-A

Abstract

The invention discloses a motion control method, boxing control method and system of a humanoid robot, and belongs to the technical field of motion control of humanoid robots. In the existing robot motion recognition scheme, collected motion data is missing and cannot be applied to motion control of a humanoid robot. The motion control method of the humanoid robot is based on the motion scene of the humanoid robot, establishes sensing topological position information, further can be used for respectively arranging one or more sensors at a plurality of positions on the motion capture clothes to form a contact type motion capture structure, acquires the motion data of an operator by utilizing the contact type motion capture structure, can effectively avoid the problem of missing motion data caused by shielding or light interference, and generates a motion control instruction about the humanoid robot based on a motion capture control algorithm, so that the robot can be accurately controlled, the effective interaction between the operator and the humanoid robot is realized, and the method is suitable for various scenes.

Inventors

  • Request for anonymity
  • Request for anonymity

Assignees

  • 宇树科技股份有限公司

Dates

Publication Date
20260508
Application Date
20260130
Priority Date
20251106

Claims (13)

  1. 1. A motion control method of a humanoid robot is characterized in that: The method comprises the following steps: Firstly, based on a previously created sensing layout model, based on a humanoid robot action scene, establishing sensing topological position information so as to respectively set one or more sensors at a plurality of positions on the action capturing garment for forming a contact action capturing structure; Step two, acquiring operator action data based on sensing topological position information through a pre-established action capturing model; Thirdly, performing kinematic calculation and data fusion on the action data of an operator by adopting an action coupling model which is created in advance to obtain multi-node fusion information; And step four, using a previously created instruction generation model, and processing the multi-node fusion information based on an action capture control algorithm to generate an action control instruction about the humanoid robot.
  2. 2. The motion control method of a humanoid robot according to claim 1, characterized in that: Step one, based on a previously created sensing layout model and based on a humanoid robot action scene, the method for establishing sensing topological position information comprises the following steps: step 11, according to the action scene of the humanoid robot, the action characteristics of the humanoid robot are obtained, wherein the action characteristics comprise shoulder action, elbow action, wrist action, trunk action, knee action and ankle action; step 12, processing the action characteristics of the humanoid robot to ensure the variables to be controlled of the humanoid robot; step 13, designing the assembly positions and the number of the sensors based on the variables to be controlled and the structural characteristics of the humanoid robot; And 14, establishing sensing topological position information according to the assembly positions and the number of the sensors, and respectively setting one or more sensors at a plurality of positions on the motion capture clothing according to the sensing topological position information to form a contact type motion capture structure.
  3. 3. A motion control method of a humanoid robot according to claim 2, characterized in that: step 13, based on the variables to be controlled and the structural characteristics of the humanoid robot, the method for designing the assembly positions and the number of the sensors is as follows: Acquiring variables to be controlled, wherein the variables comprise shoulder action variables, elbow action variables, wrist action variables, trunk action variables, knee action variables and ankle action variables to be controlled; according to the structural characteristics of humanoid robot and shoulder action variable to be controlled, inertial measurement units capable of covering front position, middle position and rear position of deltoid muscle are arranged, the number of inertial measurement units is The device is used for detecting the bending and stretching angle, the folding and unfolding angle and the rotating angle of the shoulder joint; According to the elbow motion variable to be controlled, setting the device capable of covering the inner side and the outer side of the elbow The inertial measurement unit is used for monitoring the flexion and extension postures and the supination posture of the elbow; according to the wrist motion variable to be controlled, setting the device capable of covering the palm back side The inertia measurement units are used for capturing the boxing wrist moment and the explosive force direction; Setting the covering vertebrae according to the trunk action variable to be controlled The inertia measurement units are used for detecting the lumbar vertebra bending and stretching posture and the thoracic vertebra torsion posture; according to the knee action variable to be controlled, the knee joint capable of covering the lateral ankle of femur and tibia is arranged The inertia measurement units are used for detecting acceleration of the knee joint Qu Shenjiao during leg sweeping actions in real time; According to the ankle action variable to be controlled, a tendon group capable of being embedded in the inner and outer sides of the ankle is arranged An inertial measurement unit for acquiring plantar pressure The assembly position of the individual sensors.
  4. 4. A motion control method of a humanoid robot according to claim 3, characterized in that: step 14, according to the assembly positions and the number of the sensors, the method for establishing the sensing topological position information is as follows: Step 141, obtaining The assembly position of the individual sensors; step 142, according to the theoretical degree of freedom information, for Performing degree-of-freedom redundancy check on the assembly positions and the distribution number of the individual sensors to obtain a check result; Resetting the assembly position and the distribution number of the sensor when the verification fails; after the verification is passed, simplifying the humanoid robot body into a multi-link mechanism, calculating the joint angle based on the measurement data of the inertial measurement unit, comparing with the optical dynamic capture calibration true value to obtain joint error, and executing step 143; step 143, performing minimization treatment on joint errors through a degree of freedom mapping algorithm, and obtaining a dynamic simulation result according to the wiring cost weight; When the dynamic simulation result is that the simulation does not pass, the assembly position and the distribution number of the sensor are adjusted, and the dynamic simulation is carried out again; when the dynamic simulation result is that the simulation passes, the method The assembly position and the distribution number of the individual sensors are used as sensing topological position information; the judgment standard of the simulation passing is that a redundant inertial measurement unit is deployed in a shoulder high stress area and a knee high stress area.
  5. 5. The motion control method of a humanoid robot according to claim 1, characterized in that: step two, through the motion capture model created in advance, based on the topological position information of sensing, the method for obtaining the operator motion data is as follows: Step 21, based on the sensing topological position information, deploying a corresponding inertial measurement unit on the motion capture clothing to obtain the designed motion capture clothing; Step 22, aligning the time stamps of all the inertial measurement units on the motion capture garment according to a pulse synchronization mechanism; The method comprises the steps that an operator wears motion capturing clothing, original motion data of the operator are collected in a contact mode, and the original motion data are collected by original data of a plurality of inertial measurement units; Step 23, compensating the original action data by using a temperature drift coupling compensation model to obtain a local coordinate system quaternion; step 24, converting the quaternion of the local coordinate system into the quaternion of the global coordinate system according to the skeleton chain reference system; and step 25, fusing adjacent inertial measurement unit data, and suppressing the noise of the quaternion of the global coordinate system to obtain operator action data.
  6. 6. The motion control method of a humanoid robot according to claim 5, wherein: step 22, the method for aligning the time stamps of all the inertial measurement units on the motion capture garment according to the pulse synchronization mechanism is as follows: Step 221, obtaining an initial time reference of an inertial measurement unit; Step 222, calculating an alignment time stamp of each inertial measurement unit according to the adoption period and the initial time reference, and linking the alignment time stamps to all the inertial measurement units through a wire harness; Step 223, generating a 1Hz pulse signal by using a main control board, triggering a reset local timer of the inertial measurement unit at the rising edge, and eliminating clock accumulated errors so as to align the time stamps of all the inertial measurement units; or/and, step 23, using a temperature drift coupling compensation model to compensate the original motion data, and obtaining the quaternion of the local coordinate system by the method of: step 231, acquiring the original action data measured by the inertial measurement units, and sampling the chip temperature of each inertial measurement unit in real time; step 232, calculating the temperature change rate according to the chip temperature and the sampling frequency; step 233, determining a temperature drift rate coefficient based on the temperature change rate and the temperature drift suppression requirement; and 234, performing coupling compensation on the original motion data according to the temperature drift rate coefficient and the dynamic weight to obtain a quaternion of the local coordinate system.
  7. 7. The motion control method of a humanoid robot according to claim 1, characterized in that: thirdly, adopting a pre-established action coupling model to perform kinematic calculation and data fusion on the action data of an operator, and obtaining multi-node fusion information by the following method: Step 31, acquiring operator action data, which includes joint quaternion gestures; Step 32, mapping the joint quaternion gesture into algebraic vectors in the 3-dimensional special orthogonal group to obtain a rotational degree of freedom vector; acquiring joint acceleration based on the joint quaternion gesture and the gravity component; double integration is carried out on the joint acceleration to obtain a translational degree of freedom vector; acquiring an actual rotation angle of the joint according to the quaternion gesture of the joint; Step 33, introducing S-shaped inhibition function constraint according to the actual rotation angle of the joint, the physiological limit angle of the joint and the attenuation slope parameter, and establishing a joint limit function; Step 34, correcting the rotational degree of freedom vector based on the joint limiting function, namely, when the rotational angle exceeds the limit, weighting and attenuating the rotational component to obtain a corrected rotational degree of freedom vector; Step 35, combining the corrected rotational degree of freedom vector and translational degree of freedom vector to obtain a new joint degree of freedom vector, and arranging the new joint degree of freedom vector smoothly to obtain a remodelling degree of freedom vector; Step 36, based on the joint type label, the joint type label is divided into a shoulder label, an elbow label or a wrist label, joint type embedding and space position coding are carried out on the plastic reconstruction degree of freedom vector, and multi-node fusion information is obtained.
  8. 8. The motion control method of a humanoid robot according to claim 1, characterized in that: And step four, using a previously created instruction generation model, and processing multi-node fusion information based on a dynamic capture control algorithm, wherein the method for generating the action control instruction of the humanoid robot comprises the following steps: Step 41, processing the multi-node fusion information based on a multi-head attention mechanism to generate a dynamic track; step 42, calculating the similarity between the dynamic track and a preset action template through a dynamic time warping algorithm; When the similarity is larger than the similarity threshold, taking the preset action template as an action control instruction of the humanoid robot to realize a series of action control of the humanoid robot; When the similarity is smaller than the similarity threshold, the dynamic track is processed to generate an action control instruction, and the real-time action control instruction is synchronized into the humanoid robot body by utilizing a communication protocol, so that the action control of the humanoid robot is realized.
  9. 9. The motion control method of a humanoid robot according to claim 8, characterized in that: step 41, based on the multi-head attention mechanism, the multi-node fusion information is processed, and the method for generating the dynamic track is as follows: step 411, constructing a spatio-temporal dual stream transform architecture, comprising a spatial encoder and a temporal encoder; The space encoder projects the weight matrix, calculates the inter-joint correlation weight, and then processes the multi-node fusion information based on a multi-head attention mechanism to obtain space embedding capable of representing key actions of joints; the time encoder introduces 3D rotation position embedding based on joint flexibility coefficient and action angular frequency parameter on the basis of space embedding to obtain joint rotation quantity; Step 412, converting the joint rotation amount into a bone end trajectory, i.e. a dynamic trajectory, in combination with the inverse kinematics model, while satisfying the physiological constraint and the motion smoothness requirements of the human bone chain, comprising the following contents: Step 4121, obtaining the joint rotation output by the time encoder, and constructing a bone chain constraint model from the hip joint to the wrist end according to the bone length; Step 4122, iteratively solving the end trajectory by using jacobian pseudo-inverse method based on the skeletal chain constraint model, the error vector of the joint rotation amount, and the damping coefficient: Step 4123, dynamically optimizing the tail end track according to the joint torque constraint and the reference joint angle to obtain a tail end optimized track; in step 4124, a cubic spline interpolation is applied to the end optimization trajectory to generate a continuously differentiable bone end trajectory, ensuring that the velocity and acceleration are continuous.
  10. 10. The motion control method of a humanoid robot according to claim 8, characterized in that: In step 42, the method for calculating the similarity between the dynamic track and the preset action template by the dynamic time warping algorithm is as follows: Step 421, obtaining one or more preset action templates in a preset action library, wherein the one or more preset action templates at least comprise a straight punch template, an upper hook punch template and a side kick template; acquiring a real-time action sequence of a dynamic track and a preset reference action sequence of a certain preset action template; Step 422, calculating Euclidean distance between each point in the real-time action sequence and the preset reference action sequence to form a distance matrix; Step 423, searching a path with the smallest accumulated distance from the distance matrix by using dynamic programming to obtain an optimal curved path; In step 424, the accumulated distance of the optimal curved path is the similarity between the two sequences, and the accumulated distance is small and the similarity is high.
  11. 11. A motion control method of a humanoid robot is characterized in that: The method comprises the following steps: based on the human-shaped robot action scene and the action capturing clothing, acquiring operator action data in a contact mode; performing kinematic calculation and data fusion on the action data of an operator to obtain multi-node fusion information; based on the dynamic capture control algorithm, the multi-node fusion information is processed, and an action control instruction about the humanoid robot is generated.
  12. 12. A boxing control method of a humanoid robot is characterized in that: The method comprises the following steps: acquiring human fight mechanical data based on boxing characteristics of the humanoid robot through a pre-established motion capture model; performing kinematic calculation and data fusion on the human combat mechanics data by adopting an action coupling model which is created in advance to obtain a dynamic track; generating a model by using a pre-established instruction, and matching the dynamic track with one or more preset action templates; When the matching is successful, the preset action template is directly used as an action control instruction of the humanoid robot, so that the humanoid robot can complete a series of preset boxing actions, and combat counterattack of the humanoid robot is realized; when the matching is unsuccessful, the dynamic track is processed, a real-time action control instruction is generated, and the real-time boxing action control of the humanoid robot is realized.
  13. 13. A motion control system of a humanoid robot is characterized in that: A motion control method of a humanoid robot using any one of claims 1 to 11 or a boxing control method of a humanoid robot using claim 12, comprising a plurality of inertial measurement units, a communication board, a main controller, and a transceiver module; one or more inertial measurement units are connected with the communication board by adopting a chained topology; after the communication board acquires the data of the inertial measurement unit, the data is summarized into a main controller for motion calculation through a bus; The main controller acquires the data of all the inertial measurement units in real time, executes an action track generation algorithm or/and a preset action recognition algorithm, and synchronizes the control instruction to the humanoid robot body through the transceiver module.

Description

Motion control method, boxing control method and boxing control system for humanoid robot Technical Field The invention relates to a motion control method, boxing control method and system of a humanoid robot, and belongs to the technical field of motion control of humanoid robots. Background The Chinese patent application (publication number: CN 120296603A) discloses a boxing action recognition method and system based on priori knowledge and multi-element time sequence classification, wherein a non-contact action capturing system is adopted to obtain skeleton point movement time sequence data, coarse granularity filtering based on unscented Kalman filtering is used for stabilizing a skeleton structure, fine granularity filtering based on kinematic priori knowledge is used for suppressing noise and abnormality at a time sequence coordinate value level, comprehensive analysis is carried out on the skeleton point movement time sequence data, and the kinematic priori knowledge and a multivariate time sequence classifier are utilized for recognizing and classifying attack and defending actions in boxing scenes. The scheme is mainly used for identifying boxing actions of athletes, but does not disclose how to apply the identified boxing actions to robot control, so that interactive control of the humanoid robot is affected. In addition, the scheme adopts the contactless motion capture system to collect the skeleton point motion time sequence data, and in the complex motions, multi-person interaction and dynamic environments of the boxer, shielding problems exist, such as winding and holding of the boxer, leaning motion, boxing table fence, even the body of surrounding people, strong ambient light and stage light can shield or interfere key skeleton points of the boxer, so that the collected skeleton point motion time sequence data have missing values, and accurate recognition of the motions of the boxer is affected, and therefore the scheme cannot be applied to motion control of a humanoid robot. Furthermore, through the existing teleoperation mode of the robot, the robot can simulate the simple action of an operator, but the situation of lack of identification of the action of the operator still exists, each action of the operator needs to be subjected to complex analysis calculation and wireless communication and then can be transmitted to the robot control module, and then the robot control module analyzes the transmitted data again to generate a control instruction so as to control the robot to complete the action. The whole teleoperation process needs a longer processing period, greatly influences the response speed of the robot, and cannot be suitable for countering more violent action scenes such as boxing matches, free combat matches, taylor matches, taekwondo matches and the like. Therefore, the existing teleoperation scheme cannot be suitable for fighting game scenes, and the humanoid robot cannot be enabled to make quick counterattack in extremely short time, so that the game effect of the humanoid robot is seriously affected. The information disclosed in this background is only for the understanding of the background of the inventive concept and therefore it may comprise information that does not form the prior art. Disclosure of Invention The invention aims to solve the problems or one of the problems, and provides a motion control method and a motion control system for a humanoid robot, which are used for establishing sensing topological position information based on a motion scene of the humanoid robot, so that one or more sensors can be respectively arranged at a plurality of positions on a motion capture garment to form a contact type motion capture structure, acquiring operator motion data by utilizing the contact type motion capture structure, thereby effectively avoiding the problem of missing of the motion data caused by shielding or light interference, then performing kinematic calculation and data fusion on the operator motion data to obtain multi-node fusion information, and processing the multi-node fusion information based on a motion capture control algorithm to generate motion control instructions about the humanoid robot, thereby accurately controlling the robot, realizing effective interaction between an operator and the humanoid robot, being applicable to various scenes and not influenced by surrounding objects and strong light. Aiming at the problems or one of the problems, the invention aims to provide a motion control method and a motion control system for a humanoid robot, which adopt contact type clothing to acquire operator motion data, so that the deployment efficiency of a motion capture system can be effectively improved, mass production is convenient, meanwhile, the motion capture can be realized only by putting on the motion capture clothing by an operator, an external camera or a fixed base station is not needed, deployment is convenient, manufacturing cost is not neede