CN-122008251-A - Self-adaptive joint control method and system for humanoid robot

CN122008251ACN 122008251 ACN122008251 ACN 122008251ACN-122008251-A

Abstract

The application relates to the technical field of robot control, in particular to a self-adaptive joint control method and system for a humanoid robot. The method comprises the steps of obtaining relative motion information, motion state information and internal mechanical state information to form comprehensive perception data, conducting contact prediction based on the comprehensive perception data to complete contact risk judgment and determine a potential contact area, responding to a contact prediction result, and sending a softening control instruction to a joint driver in the potential contact area through a quick instruction channel independent of a conventional track position control flow to enable the joint to be adjusted from a high-rigidity state to a low-rigidity state within a preset response time. The problem of safety risk and equipment damage that the accidental contact with the outside possibly causes when the humanoid robot executes a high-precision assembly task under a high-rigidity control state is solved, and the technical problem of response lag of the existing safety protection mechanism is solved.

Inventors

LI SHUANGQI
WANG JIWEN
BAI GUOCHAO
YU JIAN
MAO JIANLIANG
WANG YAN
WANG CHENGZHEN
HU MINCHAO

Assignees

上海云帆智控机器人科技有限公司
知行机器人科技（苏州）有限公司

Dates

Publication Date: 20260512
Application Date: 20260410

Claims (10)

1. The self-adaptive joint control method for the humanoid robot is characterized by being applied to a scene that the humanoid robot executes a sub-millimeter-level high-precision assembly task and at least part of joints are in a high-rigidity control state, and comprises the following steps: The method comprises the steps of obtaining relative motion information between an external object and a humanoid robot, motion state information of the humanoid robot and internal mechanical state information of a joint driving device to form comprehensive perception data, wherein the relative motion information at least comprises distance information and relative speed information of the external object and the humanoid robot, the motion state information at least comprises joint speed information and joint acceleration information, and the internal mechanical state information at least comprises joint motor current information and joint moment information; The contact prediction is carried out based on the comprehensive perception data to obtain a contact prediction result, wherein the contact prediction comprises the steps of judging the contact risk of the external object and the humanoid robot, and determining a potential contact area, wherein the contact risk is determined according to one or more of the following conditions that the distance information is smaller than a preset safe distance threshold value and the relative speed information is larger than a preset approaching speed threshold value; And responding to the contact prediction result, and sending a flexible control instruction to a joint driver in the potential contact area through a quick instruction channel independent of a conventional track position control flow, so that the joint in the potential contact area is adjusted from a high-rigidity state to a low-rigidity state within a preset response time, wherein the flexible control instruction comprises switching a joint control mode from position control to moment control or impedance control and/or limiting the maximum output moment of the joint to a preset safe moment upper limit.
2. The method of claim 1, wherein sending a compliance control command to the joint driver in the potential contact area to adjust the joint in the potential contact area from the high stiffness state to the low stiffness state within a preset response time, further comprising: Performing a restoration control of the joint from the low stiffness state to the high stiffness state in the potential contact area, wherein the restoration control comprises: Applying preset weak excitation to the joint, and collecting dynamic response data of the joint; Analyzing the dynamic response data to quantify the resistance capability and the quick recovery capability of the interior of the joint to weak excitation, thereby obtaining a quantified result; According to the quantification result, the recovery speed and the recovery path of the joint from the low-rigidity state to the high-rigidity state are adjusted; And continuously monitoring the actual motion state of the end effector of the humanoid robot in the process of recovering the joints so as to verify the guarantee effect of the recovery speed and the recovery path on the stability of the end effector, and adjusting the recovery speed and/or the recovery path when the actual motion state of the end effector does not meet the preset stability condition, wherein the end effector is used for executing a sub-millimeter-level high-precision assembly task.
3. The method of claim 2, wherein analyzing the dynamic response data to quantify the resistance to weak excitation and the rapid recovery of the interior of the joint, the quantifying comprising: performing time-frequency analysis on the dynamic response data to obtain energy spectrum density in a specific frequency range; comparing the energy spectrum density with a pre-established baseline energy spectrum density in a healthy state to identify a micro-vibration signal with a preset time-frequency characteristic; calculating the signal-to-noise ratio of the micro-vibration signal; Performing self-adaptive filtering processing on the micro-vibration signal with the signal-to-noise ratio lower than a preset signal-to-noise ratio threshold value so as to enhance the characteristics of the micro-vibration signal and obtain an enhanced micro-vibration signal; And quantifying the resistance and the quick recovery capacity of the interior of the joint to weak excitation according to the intensity and the duration of the enhanced micro-vibration signal, thereby obtaining the quantified result.
4. The adaptive joint control method according to claim 3, wherein applying a preset weak stimulus to the joint and collecting dynamic response data of the joint, comprises: recording the time sequence information and the excitation parameters of the weak excitation; Applying the weak excitation to the joint for multiple times under the same excitation parameters in a preset acquisition period, and respectively acquiring the corresponding dynamic response data; According to the time sequence information, the dynamic response data acquired for multiple times are synchronously aligned and synchronously overlapped and averaged to obtain an enhanced micro-vibration signal; Acquiring sensor data of other non-stimulated joints of the humanoid robot, wherein the other non-stimulated joints are joints which have a mechanical coupling relation with the joints and do not apply the weak excitation during the weak excitation; Constructing an adaptive noise cancellation filter based on the correlation between the sensor data of the other non-stimulated joints and the enhanced micro-vibration signal, and inputting the sensor data of the other non-stimulated joints as a reference noise signal into the adaptive noise cancellation filter to obtain a crosstalk component in the enhanced micro-vibration signal; Subtracting the crosstalk component from the enhanced micro-vibration signal to obtain a micro-vibration signal with crosstalk removed, wherein the micro-vibration signal is used as an input for analyzing the dynamic response data; Monitoring environmental noise in the humanoid robot working space in the process of acquiring the dynamic response data for a plurality of times in a preset acquisition period to obtain frequency and intensity information of the environmental noise; and adjusting the frequency of the weak excitation in the subsequent acquisition period according to the frequency and the intensity of the environmental noise, and adopting the adjusted frequency of the weak excitation in the subsequent weak excitation application and dynamic response data acquisition to enable the dynamic response data to correspond to the adjusted frequency of the weak excitation.
5. The method of claim 3, wherein performing a time-frequency analysis on the dynamic response data to obtain an energy spectral density in a specific frequency range comprises: preprocessing the dynamic response data to obtain preprocessed dynamic response data, wherein the preprocessing comprises removing direct current bias and power frequency interference; Obtaining a dynamically adjusted characteristic frequency range according to the recovery stage of the joint from the low-rigidity state to the high-rigidity state and historical damage evaluation data of the joint; and setting an analysis frequency band of the time-frequency analysis based on the dynamic adjusted characteristic frequency range, performing time-frequency analysis on the preprocessed dynamic response data, and calculating energy spectrum density based on the dynamic adjusted characteristic frequency range.
6. The method of adaptive joint control according to claim 5, further comprising, after calculating an energy spectral density based on the dynamically adjusted characteristic frequency range: acquiring time-frequency characteristics of environmental noise in the humanoid robot working space, acquiring sensor time-varying data of other non-stimulated joints of the humanoid robot during the weak excitation applied to the joints, and extracting vibration time-frequency characteristics of the sensor time-varying data of the other non-stimulated joints, wherein the other non-stimulated joints are joints which have a mechanical coupling relation with the joints and are not applied with the weak excitation during the weak excitation; monitoring the time-varying characteristics of the energy spectral density of the joint in the dynamically adjusted characteristic frequency range; Comparing the time-varying characteristic with the time-frequency characteristic of the environmental noise, and identifying a first time-varying component which meets a preset first similarity rule with the environmental noise; comparing the time-varying characteristics with the vibration time-frequency characteristics of the other non-stimulated joints, and identifying a second time-varying component which meets a preset second similarity rule with the vibration of the other non-stimulated joints; subtracting the first time-varying component and the second time-varying component from the time-varying characteristic to obtain an energy spectrum density time-varying characteristic with similar components removed; the comparing the energy spectrum density with a pre-established baseline energy spectrum density in a healthy state to identify the micro-vibration signal with a preset time-frequency characteristic comprises comparing the time-varying characteristic of the energy spectrum density with the time-varying characteristic of the baseline energy spectrum density to identify the micro-vibration signal with the preset time-frequency characteristic.
7. The adaptive joint control method according to claim 6, further comprising: Extracting multidimensional features of the energy spectrum density time-varying characteristics to obtain multidimensional feature vectors, wherein the multidimensional features at least comprise frequency features, amplitude features, phase features, bandwidth features, energy gravity center features and statistical features for representing time-varying distribution forms; Comparing the multi-dimensional feature vector with a pre-established damage mode feature library, and calculating the similarity between the multi-dimensional feature vector and the feature vector of each damage mode in the damage mode feature library, wherein the damage mode feature library comprises a plurality of damage types and the feature vectors corresponding to the damage types under different damage degrees; Identifying a best-matched damage mode according to the similarity, and outputting a preliminary damage evaluation result, wherein the preliminary damage evaluation result at least comprises the best-matched damage mode and the similarity corresponding to the best-matched damage mode; Based on the preliminary damage evaluation result, marking the micro-vibration signal corresponding to the multi-dimensional feature vector as an unknown damage mode when the similarity is lower than a preset similarity threshold; Continuously collecting unclassified damage event information from historical operation data of the humanoid robot, and extracting and classifying features of the damage event information to dynamically update the damage mode feature library; and adjusting the feature vector in the damage mode feature library according to the recovery stage of the joint from the low-rigidity state to the high-rigidity state so as to adapt to the change of damage features under different working conditions.
8. The adaptive joint control method according to claim 7, wherein the identifying a best-matching damage pattern according to the similarity and outputting a preliminary damage evaluation result includes: When the similarity of a plurality of damage modes exists in the similarity calculation result and is higher than the preset similarity threshold, and the similarity difference among the plurality of damage modes is smaller than the preset difference threshold, analyzing the feature vectors of the plurality of damage modes to obtain the difference information of the plurality of damage modes in the frequency feature, the amplitude feature, the duration feature and the statistical feature representing the time-varying distribution form, and carrying out weighted evaluation on the difference information by combining the current operation working condition of the joint to determine the best matched damage mode, wherein the current operation working condition at least comprises one or more of operation load, joint speed, joint output moment and environment temperature; When the similarity calculation result cannot meet preset distinguishing conditions to distinguish different damage modes, triggering a verification process, applying a group of detective excitation to the joint, and collecting transient response data of the joint under the action of the detective excitation; According to the determined type of the damage mode and the current running state of the joint, a preset damage evaluation rule is selected, the damage degree is primarily quantified by combining with an actual running parameter of the joint, and a primary damage evaluation result is output, wherein the actual running parameter at least comprises one or more of joint moment fluctuation, joint position deviation and temperature change trend, the primary damage evaluation result at least comprises the best matched damage mode and the similarity corresponding to the best matched damage mode, and the primary quantification result of the damage degree is used for determining the recovery speed and/or the adjustment direction of the recovery path for the recovery control of the joint from the low-rigidity state to the high-rigidity state.
9. The adaptive joint control method according to claim 8, wherein selecting a preset damage-assessment rule according to the determined type of damage pattern and the current running state of the joint, comprises: selecting a corresponding initial evaluation rule from a preset damage type-evaluation model mapping table according to the determined damage mode type; dynamically adjusting parameters of the initial evaluation rule according to the current running state of the joint to obtain an adjusted evaluation rule; inputting the actual operation parameters of the joint into the adjusted evaluation rule to obtain a damage degree index representing the damage degree; when the damage degree index exceeds the application range of a preset damage evaluation rule, triggering a self-adaptive correction flow, and collecting real-time operation data of the joint and signal characteristics for representing damage, wherein the signal characteristics comprise the energy spectrum density time-varying characteristics and/or the unique response characteristics, and correcting parameters and structures of the adjusted evaluation rule by combining historical similar damage case data to obtain the damage evaluation rule; In the damage evaluation process, the evolution trend of the damage signal characteristics is continuously monitored, and the damage degree index is corrected by combining the operation load and the environmental condition of the joint.
10. An adaptive joint control system for a humanoid robot, characterized by being applied to a scene in which the humanoid robot performs a sub-millimeter-level high-precision assembly task and at least part of joints are in a high-rigidity control state, the system comprising: The sensing data acquisition module is used for acquiring relative motion information between an external object and the humanoid robot, motion state information of the humanoid robot and internal mechanical state information of the joint driving device to form comprehensive sensing data, wherein the relative motion information at least comprises distance information and relative speed information of the external object and the humanoid robot, the motion state information at least comprises joint speed information and joint acceleration information, and the internal mechanical state information at least comprises joint motor current information and joint moment information; The contact prediction module is used for carrying out contact prediction based on the comprehensive perception data to obtain a contact prediction result, wherein the contact prediction comprises contact risk judgment on the possibility of contact between the external object and the humanoid robot and determination of a potential contact area, and the contact risk judgment determines that the contact risk exists according to one or more of the conditions that the distance information is smaller than a preset safety distance threshold value and the relative speed information is larger than a preset approaching speed threshold value; The joint compliance control module is used for responding to the contact prediction result, sending a compliance control command to a joint driver in the potential contact area through a rapid command channel independent of a conventional track position control flow, and enabling the joint in the potential contact area to be adjusted from a high-rigidity state to a low-rigidity state within a preset response time, wherein the compliance control command comprises switching a joint control mode from position control to moment control or impedance control and/or limiting the maximum output moment of the joint to a preset safety moment upper limit.

Description

Self-adaptive joint control method and system for humanoid robot Technical Field The application relates to the technical field of robot control, in particular to a self-adaptive joint control method and system for a humanoid robot. Background In the field of intelligent manufacturing equipment, humanoid robots play an increasingly important role, and they need to flexibly cope with various complex production environments, especially when sub-millimeter-level high-precision assembly tasks are performed, and the requirements on position accuracy are extremely high. To achieve such high precision, the driving device of the robot joint is generally set to a high stiffness control state to maximally secure the position stability of the robot arm end tool against external disturbances. However, in such a high stiffness control state, the prior art faces serious challenges when the humanoid robot is in unexpected physical contact with the surrounding environment or personnel. For example, during a precision assembly process, if a technician inadvertently touches a robot arm in a highly rigid state, a controller of the robot, which is primarily targeted to eliminate positional errors, will immediately issue a brute force correction command in an attempt to pull the arm back into the predetermined path. Because of the high rigidity characteristic of the joints, the strong correcting action cannot smooth external impact, but can cause severe rebound of the arms, serious safety risks are caused to technicians, and even precise workpieces clamped at the tail ends of the robots are damaged. While existing robotic systems are typically equipped internally with safety protection mechanisms based on force sensing or motor current monitoring for identifying collisions and initiating an emergency stop or entering a compliant mode, these safety mechanisms have an inherent time lag. From the detection of abnormal force by the sensor, the information processing and logic judgment, and the final control mode switching instruction, the whole process needs tens or even hundreds of milliseconds. Whereas the reaction speed of the high stiffness position controller is on the order of microseconds or milliseconds. This time difference results in a strong and dangerous bouncing action, which is dominated by the high stiffness properties, already occurring before the safety protection mechanism is accessible for intervention. In view of the above, there is a need in the art for improvements. Disclosure of Invention The application discloses a self-adaptive joint control method and a self-adaptive joint control system for a humanoid robot, which aim to solve the problems of safety risk and equipment damage possibly caused by accidental contact with the outside and the technical problem of response hysteresis of the existing safety protection mechanism when the humanoid robot performs a high-precision assembly task under a high-rigidity control state. The technical scheme of the application is as follows: In a first aspect, the application discloses a self-adaptive joint control method for a humanoid robot, which is applied to a scene that the humanoid robot executes a sub-millimeter-level high-precision assembly task and at least part of joints are in a high-rigidity control state, and comprises the following steps: The method comprises the steps of obtaining relative motion information between an external object and a humanoid robot, motion state information of the humanoid robot and internal mechanical state information of a joint driving device to form comprehensive perception data, wherein the relative motion information at least comprises distance information and relative speed information of the external object and the humanoid robot, the motion state information at least comprises joint speed information and joint acceleration information, and the internal mechanical state information at least comprises joint motor current information and joint moment information; The contact prediction is carried out based on comprehensive perception data to obtain a contact prediction result, wherein the contact prediction comprises the steps of carrying out contact risk judgment on the possibility of contact between an external object and human-shaped robot, and determining a potential contact area, wherein the contact risk judgment determines that the contact risk exists according to one or more of the conditions that the distance information is smaller than a preset safety distance threshold value and the relative speed information is larger than a preset approaching speed threshold value; And responding to a contact prediction result, and sending a softening control command to a joint driver in the potential contact area through a quick command channel independent of a conventional track position control flow, so that the joint in the potential contact area is adjusted from a high-rigidity state to a low-rigidity state within a preset respo