Multi-unit cooperative self-adaptive terrain robot hybrid control method and system

Abstract

The invention provides a multi-unit cooperative self-adaptive terrain robot hybrid control method and system, and relates to the field of specialized robots, and the method comprises the steps: obtaining the contact force, vibration acceleration, support foot displacement and ground texture images of support feet of a robot; calculating mechanical characteristics, vibration characteristics and visual characteristics; setting numerical ranges of the mechanical features, the vibration features and the visual features corresponding to the grounds with different rigidities; on the basis of the improved D-S evidence theory, the weights of the mechanical features, the vibration features and the visual features are set, so that the ground rigidity level is calculated; determining a mixing ratio according to the ground rigidity level, and performing weighted summation on the position control quantity and the force control quantity according to the mixing ratio to generate a robot single body control instruction; and after the previous robot single body executes the control instruction, the next robot single body is triggered to execute the control instruction. And the cooperation precision among the robot monomers in the multi-unit robot system is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of special robots, in particular, to a multi-unit cooperative self-adaptive terrain robot hybrid control method and system. BACKGROUND

[0002] In the field of robot technology, with the continuous development, legged robots have shown incomparable advantages in special environment operation scenarios due to their excellent adaptability to complex terrain. Especially in extreme and unstructured environment operation tasks such as rescue and disaster relief, geological exploration, interstellar exploration, the performance of robots is put forward extremely strict requirements, not only need to have strong terrain adaptability, also need to have precise cooperative control ability and rich operation function. In the prior art, the three-legged robot with single body structure has relatively simple structure design. Although this kind of robot has certain moving ability when facing complex terrain, its obstacle crossing ability is seriously limited by its own size. Due to the limitation of its own structure, it cannot cross higher or more complex obstacles, which greatly limits its operation range in extreme environment. Although the multi-unit modular robot has certain expansion ability in design, it can realize the expansion of function by increasing the module. However, in the actual application process, many serious problems are exposed. First of all, in the aspect of cooperative control, the multi-unit motion synchronization is poor, and the error is often more than 50ms, which makes the robot unable to realize precise cooperation between units when performing complex actions, seriously affecting the overall performance of the robot. Secondly, the coordination accuracy of joint motion and screw stretching is insufficient, and the deviation is greater than 0.1mm, which leads to the fact that when performing some operations with high precision requirements, the expected effect cannot be achieved. In addition, the stability control means of the existing multi-unit (multi-section) robot is relatively simple, which only relies on traditional gait planning to maintain balance. When the robot inclination angle exceeds 30°, it is easy to lose stability, especially when performing operation tasks and bearing additional load, the risk of imbalance increases greatly, which seriously threatens the operation safety. In low temperature environment, such as below-20℃, or in environment with humidity higher than 90%, the sensor is easy to fail, which leads to the decline of the sensing ability of the robot and the inability to work normally, which seriously restricts its application in extreme environment. In addition, in the process of realizing the present application, the applicant found that: in the unstructured environment operation of the legged robot, the interaction characteristics of the ground and the robot directly determine the motion stability and operation safety. Different ground stiffness (such as concrete, soil, snow, etc.) has significant difference in the reaction force and deformation response of the supporting foot of the robot, which needs to be dynamically matched through adaptive control. The existing control method is based on a single mechanical signal to judge the terrain, which is difficult to deal with complex mixed terrain, and becomes the core bottleneck restricting the terrain adaptability of the robot.

[0003] Therefore, how to optimize the control method of the multi-section robot to improve the operation stability and safety becomes a technical problem to be solved. SUMMARY

[0004] The present application aims to at least solve one of the technical problems existing in the prior art or related art, and discloses a multi-unit cooperative adaptive terrain robot hybrid control method and system, which solves the problems of low cooperation accuracy and poor terrain adaptability between individual robots in a traditional multi-unit robot system, and improves the operation stability and safety.

[0005] The first aspect of the present application discloses a multi-unit cooperative adaptive terrain robot hybrid control method, comprising: multi-source information acquisition: acquiring the contact force, vibration acceleration, support foot displacement and ground texture image of the robot support foot through the force sensor, vibration sensor, magnetic encoder and camera arranged on each section of the robot body; feature extraction: calculating the force change rate according to the change of the contact force, and calculating the ratio of the force change rate to the support foot displacement as the mechanical feature; extracting the vibration main frequency of the vibration acceleration as the vibration feature through Fourier transform; calculating the texture entropy of the ground texture image as the visual feature through the gray level co-occurrence matrix; calculating the ground stiffness: setting the numerical range of the mechanical feature, vibration feature and visual feature corresponding to different stiffness grounds; based on the improved D-S evidence theory, setting the weights of the mechanical feature, vibration feature and visual feature to calculate the ground stiffness grade; force loop control mode and position loop control mode compound: determining the compound control proportion λ according to the ground stiffness grade; using the dynamic impedance parameter as the impedance parameter of the force loop control mode; first calculating the product of the position control amount output by the position loop control mode and the compound control proportion λ, then calculating the product of the force control amount output by the force loop control mode and (1-λ), adding the two product results, and finally obtaining the control instruction executed by the robot body, the mathematical formula is referred to: control instruction = position control amount × λ + force control amount × (1-λ).

[0006] multi-unit cooperative control: triggering the execution of the control instruction of the next robot body after the execution of the control instruction of the previous robot body; the end robot body feeds back a cooperation ready signal to the starting robot after executing the control instruction, so as to perform global pose verification.

[0007] According to the multi-unit cooperative adaptive terrain robot hybrid control method disclosed by the present application, preferably, it further comprises: Optimize the impedance parameter: the ground stiffness level, the current load of the robot and the robot tilt angle are taken as the state input of the reinforcement learning model, and the impedance parameter of the robot force loop control mode is taken as the action output of the reinforcement learning model for model training, so as to obtain the trained mapping model; the impedance parameter of the force loop control mode is determined according to the trained mapping model and the real-time acquired ground stiffness level, and is taken as the dynamic impedance parameter.

[0008] According to the multi-unit cooperative adaptive terrain robot hybrid control method disclosed by the application, preferably, the method further comprises: Multi-source fault tolerance mechanism: when a sensor fails, the feature weight of other non-failed sensors is improved.

[0009] According to the multi-unit cooperative adaptive terrain robot hybrid control method disclosed by the application, preferably, the higher the ground stiffness is, the higher the proportion of the position control mode is, and the greater the value of lambda is.

[0010] According to the multi-unit cooperative adaptive terrain robot hybrid control method disclosed by the application, preferably, the ground stiffness level is specifically divided into five levels: extremely hard, hard, medium, soft and extremely soft.

[0011] The second aspect of the application discloses a multi-unit cooperative adaptive terrain robot hybrid control system, comprising: a memory for storing program instructions; a processor for calling the program instructions stored in the memory to realize the multi-unit cooperative adaptive terrain robot hybrid control method of any of the above technical solutions.

[0012] Compared with the prior art, the application has at least the following beneficial effects: multi-source sensor data fusion reduces the misjudgment rate compared with single sensor judgment, the ground stiffness is determined by the mechanical characteristics, vibration characteristics and visual characteristics, the identification accuracy of the ground stiffness is improved, the ground support characteristics are matched, and the adaptability to complex terrain is improved; the weights of the force loop control and the position loop control are determined according to the ground stiffness, the rigid ground impact is reduced, the soft ground support stability is improved, and then the operation stability and safety of the robot single body are improved; the control instructions of each robot single body are executed in sequence, the connection error between each robot unit (joint) is reduced, and the overall trajectory deviation rate of the multi-unit robot system is reduced. The multi-source fusion fault tolerance mechanism can still maintain cooperative control when a single sensor fails, and the fault-free operation time of the system is prolonged. BRIEF DESCRIPTION OF DRAWINGS

[0013] Figure 1 A flowchart of a multi-unit cooperative adaptive terrain robot hybrid control method according to an embodiment of the application is shown.

[0014] Figure 2A schematic block diagram of a multi-unit cooperative adaptive terrain robot hybrid control system according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0015] In order to more clearly understand the above-mentioned purposes, features and advantages of the present application, the present application will be further described in detail below in conjunction with the accompanying drawings and specific embodiments. In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, however, the present application can also be implemented in other ways different from those described herein, and therefore, the present application is not limited to the specific embodiments disclosed below.

[0016] As Figure 1 shown, according to an embodiment of the present application, a multi-unit cooperative adaptive terrain robot hybrid control method is disclosed, comprising: S101, multi-source information collection: the contact force, vibration acceleration, support foot displacement and ground texture image of the robot support foot are obtained through the force sensor, vibration sensor, magnetic encoder and camera arranged on each robot unit; S102, feature extraction: first, the force change rate is calculated according to the change of the contact force, and then the ratio of the force change rate to the support foot displacement is obtained as the mechanical feature. The mathematical formula is referred to: mechanical feature = force change rate / support foot displacement; the vibration main frequency of the vibration acceleration is extracted as the vibration feature through Fourier transform; the texture entropy of the ground texture image is calculated as the visual feature through the gray level co-occurrence matrix; S103, calculation of ground stiffness: the numerical range of the mechanical feature, vibration feature and visual feature corresponding to different stiffness grounds is set; the weights of the mechanical feature, vibration feature and visual feature are set based on the improved D-S evidence theory to calculate the ground stiffness grade; S104, compound of force loop control mode and position loop control mode: the compound control proportion λ is determined according to the ground stiffness grade; the dynamic impedance parameter is used as the impedance parameter of the force loop control mode (variable parameter impedance control); the control command is obtained by multiplying the position control amount output by the position loop control mode (adaptive PID) and λ, and adding the product of the force control amount output by the force loop control mode and (1-λ), so that the robot unit executes the control command; S105, multi-unit cooperative control: the control command of the next robot unit is triggered after the control command of the previous robot unit is executed; the cooperative ready signal is fed back to the starting end robot after the end robot unit executes the control command, so as to perform global pose verification.

[0017] In this embodiment, a six-axis force sensor is used to collect the vertical force of the support foot; a three-axis vibration sensor is used to collect the vibration acceleration of the support foot; a miniature camera is used to collect the ground texture image; and a magnetic encoder is used to collect the support foot displacement.

[0018] Improved D-S evidence theory is adopted: Basic probability distribution: basic weights are assigned to three features for basic probability calculation, wherein the weight of the mechanical feature is set to 0.4, the weight of the vibration feature is set to 0.3, and the weight of the visual feature is set to 0.3.

[0019] Output: ground stiffness level (5 levels: extremely hard / hard / medium / soft / extremely soft) and confidence (≥95%).

[0020] The mapping relationship is trained based on reinforcement learning (PPO algorithm): The state input of the reinforcement learning model includes three data, namely the ground stiffness level, the current load of the robot and the tilt angle of the robot; the action output of the model is a set of impedance parameters, which includes three dimensions of virtual mass (denoted by M), virtual damping (denoted by B) and virtual stiffness (denoted by K).

[0021] Parameter range after training: After the model training is completed, the impedance parameter ranges corresponding to different ground stiffness levels are as follows: when the ground is extremely hard, the virtual mass M is 0.5 kg, the virtual damping B is 30 N·s / m, and the virtual stiffness K is 1000 N / m; when the ground is extremely soft, the virtual mass M is 2 kg, the virtual damping B is 80 N·s / m, and the virtual stiffness K is 200 N / m.

[0022] According to the above embodiment, preferably, it further comprises: optimizing the impedance parameters: taking the ground stiffness level, the current load of the robot and the tilt angle of the robot as the state input of the reinforcement learning model, and taking the impedance parameters of the robot force loop control mode as the action output of the reinforcement learning model to perform model training, to obtain a trained mapping model; determining the impedance parameters of the force loop control mode as the dynamic impedance parameters according to the trained mapping model and the real-time acquired ground stiffness level.

[0023] According to the above embodiment, preferably, it further comprises: multi-source fault-tolerant mechanism: when a certain sensor fails, the feature weight of other non-failed sensors is increased.

[0024] According to the above embodiment, preferably, the higher the ground stiffness, the higher the proportion of the position control mode, and the greater the value of λ.

[0025] According to the above embodiment, preferably, the ground stiffness level is specifically divided into 5 levels: extremely hard, hard, medium, soft and extremely soft.

[0026] As Figure 2As shown, according to still another embodiment of the present application, a multi-unit cooperative adaptive terrain robot hybrid control system 200 is also disclosed, comprising: a memory 201 for storing program instructions; and a processor 202 for invoking the program instructions stored in the memory to implement the multi-unit cooperative adaptive terrain robot hybrid control method of the above-mentioned embodiments.

[0027] All or part of the steps in the various methods of the above-mentioned embodiments can be completed by a program controlling relevant hardware, which can be stored in a readable storage medium, including a Read-Only Memory (ROM), a Random Access Memory (RAM), a Programmable Read-only Memory (PROM), an Erasable Programmable Read Only Memory (EPROM), a One-time Programmable Read-Only Memory (OTPROM), an Electrically-Erasable Programmable Read-Only Memory (EEPROM), a Compact Disc Read-Only Memory (CD-ROM) or other optical disk storage, a magnetic disk storage, a magnetic tape storage, or any other readable medium capable of carrying or storing data.

[0028] The above merely provides preferred embodiments of the present application but not for limiting the present application. For those skilled in the art, the present application can have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A hybrid control method for a multi-unit cooperative adaptive terrain robot, used in a robot system composed of multiple robot units, characterized in that, include: Multi-source information acquisition: The contact force, vibration acceleration, displacement of the supporting foot, and ground texture images of the robot are acquired by force sensors, vibration sensors, magnetic encoders, and cameras installed on each robot segment; Feature extraction: The rate of change of force is calculated based on the change of contact force, and the ratio of the rate of change of force to the displacement of the supporting foot is calculated as a mechanical feature; the dominant frequency of the vibration acceleration is extracted by Fourier transform as a vibration feature; the texture entropy of the ground texture image is calculated by gray-level co-occurrence matrix as a visual feature. Calculate ground stiffness: Define the numerical range of mechanical characteristics, vibration characteristics, and visual characteristics corresponding to different ground stiffnesses; Based on the improved DS evidence theory, the weights of the mechanical characteristics, vibration characteristics and visual characteristics are set to calculate the ground stiffness level; Force loop control mode and position loop control mode combined: The combined control ratio λ is determined according to the ground stiffness level; the dynamic impedance parameter is used as the impedance parameter of the force loop control mode; first calculate the product of the position control quantity output by the position loop control mode and the combined control ratio λ, then calculate the product of the force control quantity output by the force loop control mode and (1-λ), add the two product results together, and finally obtain the control command executed by the robot. Multi-unit cooperative control: After the preceding robot executes a control command, the following robot is triggered to execute a control command; after the end-effector executes a control command, it sends a cooperative ready signal back to the starting robot for global pose verification.

2. The multi-unit cooperative adaptive terrain robot hybrid control method according to claim 1, characterized in that, Also includes: Optimize impedance parameters: Use the ground stiffness level, the robot's current load, and the robot's tilt angle as the state inputs of the reinforcement learning model, and use the impedance parameters of the robot's force loop control mode as the action outputs of the reinforcement learning model to train the model and obtain the trained mapping model; determine the impedance parameters of the force loop control mode based on the trained mapping model and the real-time acquired ground stiffness level, and use them as dynamic impedance parameters.

3. The multi-unit cooperative adaptive terrain robot hybrid control method according to claim 1, characterized in that, Also includes: Multi-source fault tolerance mechanism: When a sensor fails, the feature weights of other unfailed sensors are increased.

4. The multi-unit cooperative adaptive terrain robot hybrid control method according to claim 1, characterized in that, The higher the ground stiffness, the higher the proportion of position control mode, and the larger the value of λ.

5. The multi-unit cooperative adaptive terrain robot hybrid control method according to claim 1, characterized in that, The ground stiffness is specifically divided into 5 levels: extremely hard, hard, medium, soft, and extremely soft.

6. A hybrid control system for a multi-unit cooperative adaptive terrain robot, characterized in that, include: Memory, used to store program instructions; A processor is configured to invoke the program instructions stored in the memory to implement the multi-unit cooperative adaptive terrain robot hybrid control method as described in any one of claims 1 to 5.

Images