A method and system for cooperative control of robot swarms

Abstract

This invention relates to a robot swarm cooperative control method and system, comprising the following steps: collecting structural state information; collecting terrain information; constructing a set of action patterns; setting an action cost function; evaluating the complexity of feasible actions in the set of action patterns based on the structural state information and terrain information; selecting the action pattern or combination of action patterns with the lowest action cost function value to generate an action execution sequence; and controlling the wheel-leg motion module and gripper mechanism to complete propulsion control according to the action execution sequence. This invention achieves optimal action selection and minimal-complexity propulsion for robots in complex terrain environments by comprehensively evaluating multiple action patterns through a cost function.

Description

Technical Field

[0001] This invention relates to the field of robotics, and more specifically, to a method and system for controlling collaborative robot swarms. Background Technology

[0002] With the development of biomimetic wheeled-legged robots, bimodal locomotion schemes combining wheeled and legged motion modes have been widely applied in complex terrain environments. However, existing wheeled-legged hybrid robots mostly employ fixed gait switching strategies, failing to autonomously optimize motion sequences based on task requirements, structural state, and terrain complexity, leading to the following technical problems:

[0003] First, existing motion planning technologies are only designed to achieve motion goals, lacking a comprehensive assessment of the complexity of structural form switching, the connection status of individual robots, and posture stability. This can easily lead to unnecessary mode switching and structural reconstruction, increasing execution complexity.

[0004] Secondly, traditional methods do not incorporate the connection behaviors, auxiliary actions, and support component activation actions in modular combination structures into a unified cost evaluation system. They lack an overall balance between "structural action complexity" and "behavior execution cost," making it difficult to meet the requirements of robot swarm collaborative control systems for minimizing action sequences and ensuring stability in multiple scenarios.

[0005] Therefore, there is an urgent need to propose a motion complexity optimization propulsion control method based on cost function, which can evaluate and filter the costs of multiple motion modes based on structural state, terrain information and task type, generate the execution sequence with the lowest motion complexity, reduce unnecessary structural deformation and combination switching, and improve the execution efficiency and stability of the robot in complex terrain. Summary of the Invention

[0006] This research aims to provide a collaborative control system for robot swarms that combines wheeled and legged motion modes for dual-modal movement and achieves low-cost control, applicable to complex terrain environments.

[0007] It includes a main control module, a measurement module, and multiple interconnected individual robots. Each individual robot includes a torso module, a front connecting arm, and a wheel-leg motion module.

[0008] One end of the front connecting arm is connected to the torso module via a multi-degree-of-freedom joint, and the other end is connected to a gripper mechanism. The gripper mechanism is equipped with a pressure sensor, a mechanical docking mechanism, and a docking electrode.

[0009] The wheel-leg motion module includes a hip joint module, a thigh joint module, a lower leg joint module, and a wheel foot module; there are two sets of wheel-leg motion modules, located on both sides of the torso module;

[0010] The wheel-leg motion module is used to execute wheel rolling mode and leg gait mode, the gripper mechanism is used to execute gripper support mode, and also to execute the connection or separation mode of individual robots, the measurement module is used to collect structural state information and terrain information, and the main control module is used to perform motion cost function calculation and generate motion execution sequence, and control one or more individual robots through motion execution sequence.

[0011] The present invention also provides a robot swarm cooperative control method, based on the aforementioned robot swarm cooperative control system, comprising the following steps:

[0012] Collect structural status information, which includes the deployment status of the wheel-leg motion module, the number of connected individual robots, the activation status of the gripper mechanism, and the current posture parameters;

[0013] Collect terrain information, including terrain type, obstacle size, and path accessibility level;

[0014] Construct a set of action modes, which includes wheel rolling mode, leg gait mode, gripper support mode, single robot connection mode and single robot separation mode;

[0015] Define an action cost function, which is used to evaluate the execution cost of an action pattern;

[0016] Based on structural state information and terrain information, the complexity of actionable actions in the action pattern set is evaluated according to the action cost function.

[0017] Select the action pattern or combination of action patterns with the lowest action cost function value to generate an action execution sequence;

[0018] The wheel-leg motion module and gripper mechanism are controlled according to the action execution sequence.

[0019] Preferably, when evaluating the complexity of possible actions in the action pattern set based on structural state information and terrain information, according to the action cost function:

[0020] Candidate actions in the action pattern set are filtered based on the path access level, and action patterns that do not match the path access level are excluded.

[0021] The complexity of the remaining candidate action patterns is evaluated based on the action cost function;

[0022] Select the action pattern or combination of action patterns with the lowest action cost function value from all action patterns that meet the filtering criteria.

[0023] Preferably, when evaluating the complexity of the remaining candidate action patterns based on the action cost function:

[0024] Obtain the unit action energy consumption, number of structural deformations, action execution duration, and attitude stability change value for each candidate action pattern;

[0025] The attitude stability change value is nonlinearly amplified, and the squared attitude stability change value is used as the stability score to improve the sensitivity to stability differences.

[0026] Calculate the logarithm of the energy consumption per unit motion to the base of the natural constant e, and multiply the result by the weight of the energy consumption per unit motion calculation to generate a score for the energy consumption per unit motion.

[0027] The weight is calculated by multiplying the number of structural deformations by the number of structural deformations to generate a structural deformation score.

[0028] The action execution time is exponentially amplified by multiplying the exponential value of the action execution time by the action execution time calculation weight to generate an action execution time score.

[0029] The cost function value of each candidate action mode is obtained by weighted summing of stability score, unit action energy consumption score, structural deformation number score and action execution time score.

[0030] Compare the cost function values ​​of all candidate action patterns, select the action pattern or combination of action patterns with the smallest cost function value, and generate the action execution sequence.

[0031] Preferably, the energy consumption per unit motion is calculated based on the integral of the product of the current and voltage of the motor built into the wheel-leg motion module during the motion execution cycle;

[0032] The number of structural deformations is the number of joint-driven actions during the transition from the current structural state to the target action mode. Joint drives include unfolding, contraction, and rotation, and each occurrence is recorded as 1.

[0033] The execution duration of an action is calculated using the predefined trajectory execution cycle of the action pattern;

[0034] The attitude stability change value is calculated by the shift in the center of gravity position and the change in attitude angle before and after the execution of the action mode.

[0035] Preferably, when selecting the action pattern or combination of action patterns with the lowest action cost function value to generate the action execution sequence:

[0036] If the action execution sequence includes a wheel rolling mode, the control wheel leg motion module drives the wheel set to roll at a constant speed while keeping the joint in a locked state;

[0037] If the action execution sequence includes a foot gait mode, the hip joint module, thigh joint module, and calf joint module of the control wheel leg motion module will complete the swing and support transition according to the preset gait timing.

[0038] If the action execution sequence includes a gripper support mode, then control the gripper mechanism to extend downward and keep it in contact with the ground, serving as an additional support point for balance control;

[0039] If the action execution sequence includes a single robot connection mode or a single robot separation mode, the control of the front connecting arm drives the gripper mechanism to perform mechanical docking or release operations, and detects whether the docking electrode is conductive, and detects whether the pressure reaches the preset value through the pressure sensor to determine whether the mechanical docking is successful.

[0040] Preferably, when generating an action execution sequence, a directed graph is constructed with action patterns as graph nodes and action cost function values ​​as edge weights. A graph search algorithm is used to search for the path with the lowest cumulative cost from the initial state to the target state, and the action nodes in the path are added to the action execution sequence in sequence.

[0041] Preferably, the main control module determines whether it is necessary to separate individual robots based on the task type. If it is necessary to separate individual robots, after separation, the main control module assigns corresponding target positions or sub-tasks to each individual robot based on the task target list and environmental information. Each separated individual robot independently generates an action execution sequence adapted to its task target.

[0042] Preferably, when performing an action that includes a single-robot connection mode or a single-robot separation mode, the following judgment is performed:

[0043] When the target action is a single robot connection mode, it is determined whether the remaining power of the target single robot to be connected is higher than the first power threshold and whether the communication link is in a stable connection state. If either condition is not met, the target single robot connection action is removed from the candidate actions.

[0044] When the target action is in module separation mode, it is determined whether there are any abnormal markers in the historical fault records of the target robot to be separated. If there are abnormal markers, the current combination state is retained and the separation action of the robot is paused.

[0045] Preferably, the calculation weights of the action cost function are dynamically adjusted according to the task type and terrain complexity, wherein:

[0046] When the task type is a high-speed movement task, the weight of the unit action energy consumption calculation is less than the weight of the action execution time calculation.

[0047] When the task type is a stable obstacle crossing task, the weight value of the attitude stability change value is greater than the weight value of the unit motion energy consumption.

[0048] When the detected terrain complexity exceeds the preset terrain complexity threshold, the weight of the number of structural deformations is adjusted to 1.5 times the current weight value.

[0049] The beneficial effects of this invention are as follows: By constructing an action cost function that includes unit action energy consumption, number of structural deformations, action execution time, and posture stability change values, this invention achieves a comprehensive quantitative evaluation of the execution cost of multiple action modes; by evaluating the complexity of wheeled rolling mode, legged gait mode, gripper support mode, and single robot connection or separation mode through the action cost function, it can simultaneously consider motion energy consumption, structural switching complexity, execution time, and stability during the action mode selection process; by setting structural action screening rules in the action mode screening stage, unnecessary structural deformation and gripper mechanism activation actions are reduced, thereby reducing structural disturbances and loads during action execution; by using a graph search-based path planning algorithm combined with the action cost function to generate an action execution sequence with minimum action complexity, efficient propulsion control of the robot in complex terrain environments is achieved; through this method, the robot swarm cooperative control system can dynamically adjust the weight of the action cost function according to the task type and terrain complexity, realize a task-oriented action selection strategy, improve multi-scenario adaptability, and enable the robot to dynamically adjust the connection or separation state of single robots according to the task scenario to meet the needs of different tasks for search coverage, propulsion efficiency, and workload capacity. Attached Figure Description

[0050] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the present invention will be further described below in conjunction with the accompanying drawings and embodiments. The drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0051] Figure 1 This is a flowchart of a preferred embodiment of the robot swarm cooperative control method of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, a clear and complete description will be provided below in conjunction with the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0053] It includes a main control module, a measurement module, and multiple interconnected individual robots. Each individual robot includes a torso module, a front connecting arm, and a wheel-leg motion module.

[0054] One end of the front connecting arm is connected to the torso module via a multi-degree-of-freedom joint, and the other end is connected to a gripper mechanism. The gripper mechanism is equipped with a pressure sensor, a mechanical docking mechanism, and a docking electrode. The mechanical docking mechanism is specifically a toothed engagement mechanism.

[0055] The wheel-leg motion module includes a hip joint module, a thigh joint module, a lower leg joint module, and a wheel foot module; there are two sets of wheel-leg motion modules, located on both sides of the torso module;

[0056] The wheel-leg motion module is used to execute wheeled rolling mode and legged gait mode; the gripper mechanism is used to execute gripper support mode and also to execute the connection or separation mode of individual robots; the measurement module is used to collect structural state information and terrain information; and the main control module is used to calculate motion cost function and generate motion execution sequence, which controls one or more individual robots. The measurement module includes an inertial measurement unit, LiDAR, and depth camera.

[0057] The specific workflow is as follows:

[0058] Collect structural status information. The main control module first calls the sensors of each module to obtain the deployment status of the wheel leg motion module (e.g., both wheel leg modules are currently in the wheel deployment state), the number of connected units of a single robot (single mode), the activation status of the gripper mechanism (gripper retraction is not enabled), and the attitude angle and center of gravity position output by the inertial measurement unit.

[0059] Collect terrain information. The robot uses LiDAR to scan the point cloud within a 5-meter range in front, uses a depth camera to obtain the terrain contour in the direction of crossing, and combines the detected current ground tilt angle of 5° to determine the terrain type as "loose rubble + slight slope", the maximum obstacle size is 15cm, and the path passability level is determined to be "medium-level feasible area".

[0060] Build a collection of action patterns. Generate a collection containing the following action patterns:

[0061] M1: Wheel-type scrolling mode;

[0062] M2: Foot-based gait pattern;

[0063] M3: Gripper support mode;

[0064] M4: Single robot connection mode;

[0065] M5: Single robot separation mode.

[0066] Define the action cost function. Assign initial weights to each parameter of the action cost function:

[0067] Energy consumption weight per unit action: 1;

[0068] Structural deformation number weight: 2;

[0069] Action execution duration weight: 1;

[0070] Weight of attitude stability change value: 3.

[0071] Since the current mission is a rubble search and rescue mission, and the goal is to cross obstacles with high stability, the weight of the attitude stability change value is increased to 4 according to the mission type, while the weights of other parameters remain unchanged.

[0072] The main control module first filters the action pattern set, excluding single-robot connection mode and single-robot separation mode, because the current number of single-robot connections is single-mode and the path width is insufficient for combination. Then, it detects the cumulative activation count of gripper support mode and auxiliary support component assist mode in the past 60 seconds. The gripper support mode has been used 3 times in the window, exceeding the threshold of 2 times, so the gripper support mode is removed from the candidate actions. The final candidate action modes are M1 (wheel rolling mode), M2 (leg gait mode), and M3 (gripper support mode).

[0073] The complexity of the three action patterns is evaluated based on the action cost function, and the calculation results are as follows:

[0074] M1: Total cost = 4;

[0075] M2: Total cost = 6;

[0076] M3: Total cost value = 8.

[0077] S6. Generate action execution sequence. Since the wheel rolling mode has the lowest cost, M1 is used as the action execution unit for the current stage. The graph search algorithm is used to predict the terrain conditions of the subsequent area and generate a combined action sequence [M1, M1, M2] in advance. After rolling twice in flat areas, the sequence switches to foot gait to cross obstacles with a height greater than 20cm in front.

[0078] Perform motion control. Control each wheel and leg motion module and gripper mechanism according to the generated motion execution sequence:

[0079] When executing M1, the drive wheel and foot module motor rolls forward at a speed of 0.8 m / s, keeping the hip joint module, thigh joint module, and calf joint module locked.

[0080] When an obstacle 20cm ahead is detected and the mode is switched to M2, the hip joint module is controlled to pitch to 45°, the thigh joint module to bend to 90°, and the lower leg joint module is adjusted to the crossing angle to achieve the foot gait mode to cross the obstacle.

[0081] Through the above process, the robot only performs necessary actions in the ruins, avoiding unnecessary activation of the gripper mechanism, reducing structural deformation and energy consumption, and improving crossing stability and task execution efficiency.

[0082] In this embodiment, the modular bionic wheeled robot has the function of dynamically adjusting the connection or separation state of individual robots according to the task scenario, so as to meet the requirements of different tasks for search coverage, propulsion efficiency and workload capacity. The main control module first determines the current execution target based on the task type information provided by the task allocation module.

[0083] When the task type is a distributed area search task, the main control module determines that it needs to cover a larger area in a short time to improve search efficiency. Therefore, it lowers the weight of the cost function calculation of the "single robot separation mode" in the action mode set to reduce the cost function evaluation value of the separation action. When generating the action execution sequence, it prioritizes the selection of actions containing the single robot separation mode and controls each single robot to perform the separation operation. The separation operation includes controlling the front connecting arm to drive the gripper mechanism to release the connection state, ensuring that the single robots can move independently after being separated from each other.

[0084] When the distributed search task is completed and the task type switches to heavy-load operation or high-speed propulsion task, the main control module determines that it is necessary to improve the overall load capacity of the robot or stabilize the propulsion efficiency. Therefore, the weight of the action cost function calculation of the "single robot connection mode" in the action mode set is reduced to lower the cost function evaluation value of the connection action. When generating the action execution sequence, actions containing the single robot connection mode are selected first to control each single robot to perform the connection operation. The connection operation includes controlling the front connecting arm to drive the gripper mechanism to complete the mechanical docking, and detecting the connection status through pressure sensors and docking electrodes to ensure that a stable combined robot structure is formed after the module is successfully connected.

[0085] Through the above strategy, the robot can switch between decentralized search and centralized operation configuration according to the dynamic changes of the task objectives, taking into account both area coverage and operation capability, forming a task-oriented adaptive configuration global control scheme.

[0086] The preferred embodiment of the robot swarm cooperative control method of the present invention is as follows: Figure 1 As shown, it includes the following steps:

[0087] Collect structural status information, which includes the deployment status of the wheel-leg motion module, the number of connected individual robots, the activation status of the gripper mechanism, and the current posture parameters;

[0088] Collect terrain information, including terrain type, obstacle size, and path accessibility level;

[0089] Construct a set of action modes, which includes wheel rolling mode, leg gait mode, gripper support mode, single robot connection mode and single robot separation mode;

[0090] Define an action cost function, which is used to evaluate the execution cost of an action pattern;

[0091] Based on structural state information and terrain information, the complexity of actionable actions in the action pattern set is evaluated according to the action cost function.

[0092] Select the action pattern or combination of action patterns with the lowest action cost function value to generate an action execution sequence;

[0093] The wheel-leg motion module and gripper mechanism are controlled according to the action execution sequence.

[0094] In this embodiment, when evaluating the complexity of possible actions in the action pattern set based on structural state information and terrain information according to the action cost function:

[0095] Candidate actions in the action pattern set are filtered based on the path access level, and action patterns that do not match the path access level are excluded.

[0096] The complexity of the remaining candidate action patterns is evaluated based on the action cost function;

[0097] Select the action pattern or combination of action patterns with the lowest action cost function value from all action patterns that meet the filtering criteria.

[0098] When evaluating the complexity of the remaining candidate action patterns based on the action cost function:

[0099] Obtain the unit action energy consumption, number of structural deformations, action execution duration, and attitude stability change value for each candidate action pattern;

[0100] The attitude stability change value is nonlinearly amplified, and the squared attitude stability change value is used as the stability score to improve the sensitivity to stability differences.

[0101] Calculate the logarithm of the energy consumption per unit motion to the base of the natural constant e, and multiply the result by the weight of the energy consumption per unit motion calculation to generate a score for the energy consumption per unit motion.

[0102] The weight is calculated by multiplying the number of structural deformations by the number of structural deformations to generate a structural deformation score.

[0103] The action execution time is exponentially amplified by multiplying the exponential value of the action execution time by the action execution time calculation weight to generate an action execution time score.

[0104] The cost function value of each candidate action mode is obtained by weighted summing of stability score, unit action energy consumption score, structural deformation number score and action execution time score.

[0105] Compare the cost function values ​​of all candidate action patterns, select the action pattern or combination of action patterns with the smallest cost function value, and generate the action execution sequence.

[0106] When constructing an action pattern set, the action pattern set is represented as follows:

[0107]

[0108] Each action mode Corresponding executable pass level set The current route access level is ;

[0109] When filtering candidate actions in the action pattern set and excluding action patterns that do not match the path accessibility level, the action filtering criteria are defined as follows:

[0110]

[0111] in, This is the set of filtered action patterns;

[0112] The stability score is represented as:

[0113]

[0114] The energy consumption per unit of motion is rated as follows:

[0115]

[0116] The score for the number of structural deformations is expressed as:

[0117]

[0118] The action execution time score is expressed as follows:

[0119]

[0120] in, This represents the exponential value of the action execution time, with the natural constant e as the base.

[0121] These are the weights for unit motion energy consumption, number of structural deformations, motion execution time, and posture stability.

[0122] After screening, The complexity of all action patterns is evaluated based on the action cost function, and the cost value of each action pattern is calculated. The action cost function is expressed as:

[0123]

[0124] Right now:

[0125]

[0126] in: Energy consumption per unit of action; The number of structural deformations; Duration of action execution; This represents the change in attitude stability. These are the calculated weights for each action mode;

[0127] Select cost function value The smallest combination of action patterns is used as the action to be executed in the current advancement phase.

[0128] Before generating the action execution sequence, the main control module performs an action pattern filtering step. This step first filters candidate actions in the action pattern set based on path traversability level. The path traversability level is generated by the terrain information collected in step S2, and calculated by combining the LiDAR point cloud density distribution, depth camera visual gradient, and robot attitude angle 3D tilt value. The traversability level can be divided into "high-level feasible area", "medium-level feasible area", "low-level feasible area" and "infeasible area".

[0129] During the filtering process, if a candidate action mode does not match the terrain accessibility level (for example, the wheel rolling mode may cause structural jamming when executed in a "low-level feasible area"), then the action mode will be removed from the candidate set.

[0130] Through screening and evaluation processes, infeasible or high-risk action modes can be eliminated before action mode selection, avoiding unnecessary complex action calculations, reducing computational load, improving the rationality and stability of action execution sequences, and meeting the task execution requirements of robots in complex and ever-changing environments.

[0131] Energy consumption per unit of motion Driven by motor current With voltage The integral calculation within the action execution cycle T is expressed as:

[0132]

[0133] Number of structural deformations This represents the number of joint-driven actions during the transition from the current structural state to the target action mode. Joint drives include expansion, contraction, and rotation, with each occurrence counted as 1. The joints include the hip joint module, thigh joint module, and calf joint module.

[0134] Action execution time Calculate the execution cycle based on the predefined trajectory of the action pattern;

[0135] Attitude stability change value P By shifting the center of gravity position before and after the action pattern is executed , , and attitude angle changes , , Calculate the stability change value;

[0136] Represented as:

[0137]

[0138] Where k is the attitude angle change weighting coefficient.

[0139] By incorporating four parameters—unit motion energy consumption, number of structural deformations, motion execution time, and attitude stability change—into the same cost function, and dynamically adjusting the calculation weights of each parameter based on task type and terrain complexity, the motion mode selection process not only considers minimizing energy consumption but also takes into account structural motion complexity, execution time efficiency, and attitude stability, forming a multi-objective trade-off optimization. Furthermore, it avoids the unnecessary structural switching or stability degradation problems caused by traditional optimization methods that only use energy consumption or path length as objective functions. Simultaneously, it improves the overall performance of the robot swarm cooperative control system in complex terrain tasks, enhancing the intelligence and practicality of the propulsion control strategy.

[0140] This cost function design introduces "structural motion complexity" and "attitude stability change value" into the robot motion mode evaluation system for the first time. Combined with traditional energy consumption and execution time calculation, it forms a special motion optimization method for the structural characteristics of robot swarm collaborative control system.

[0141] In this embodiment, the main control module determines whether it is necessary to separate individual robots based on the task type. If it is necessary to separate individual robots, after separation, the main control module assigns corresponding target positions or sub-tasks to each individual robot according to the task target list and environmental information. Each separated individual robot independently generates an action execution sequence adapted to its task target.

[0142] Specifically, when performing a search task in a dispersed area, the main control module controls each individual robot to separate, forming multiple sub-robots that can act independently, in order to improve the search coverage and the efficiency of parallel operations.

[0143] The main control module determines the task as a distributed search task based on the task type and performs the following operations: calls the action cost function to reduce the weight of the "single robot separation mode"; generates an action execution sequence containing the separation action mode; and controls the front connecting arm and gripper mechanism to perform a mechanical release operation to complete the separation of the single robot.

[0144] After separation, the main control module assigns different target locations or sub-tasks to each individual robot based on the task target list and environmental information, including but not limited to: environmental map construction of target area A; obstacle size measurement of target area B; and passable path evaluation of target area C.

[0145] The main control module independently generates an action execution sequence adapted to the task objectives of each separated individual robot. Each individual robot completes the target task according to the generated action execution sequence and sends the task execution status, perception data and structural status information back to the main control module in real time. The main control module then redistributes subsequent task objectives based on the task completion status of each individual robot.

[0146] Determine if the module connection conditions are met. If they are met, generate the execution sequence of the connection action mode, control each individual robot to complete the mechanical connection, restore the combination mode, and enter the subsequent propulsion operation stage.

[0147] This embodiment generates independent action execution sequences for the separated individual robots based on their respective task objectives, realizing adaptive dynamic configuration control of modular bionic wheeled robots in multi-objective tasks such as distributed search, environmental mapping, obstacle detection, and path planning, significantly improving task completion efficiency and system environmental adaptability.

[0148] In this embodiment, when selecting the action pattern or combination of action patterns with the lowest action cost function value to generate the action execution sequence:

[0149] If the action execution sequence includes a wheel rolling mode, the control wheel leg motion module drives the wheel set to roll at a constant speed while keeping the joint in a locked state;

[0150] If the action execution sequence includes a foot gait mode, the hip joint module, thigh joint module, and calf joint module of the control wheel leg motion module will complete the swing and support transition according to the preset gait timing.

[0151] If the action execution sequence includes a gripper support mode, then control the gripper mechanism to extend downward and keep it in contact with the ground, serving as an additional support point for balance control;

[0152] If the action execution sequence includes a single robot connection mode or a single robot separation mode, the control of the front connecting arm drives the gripper mechanism to perform mechanical docking or release operations, and detects whether the docking electrode is conductive, and detects whether the pressure reaches the preset value through the pressure sensor to determine whether the mechanical docking is successful.

[0153] When the current action mode in the action execution sequence is the gripper support mode, the main control module controls the gripper mechanism to extend downward in the vertical direction until the gripper pressure sensor detects a contact signal, maintaining the position of the gripper mechanism, providing additional support, and improving the robot's lateral or longitudinal stability.

[0154] When the current action mode in the action execution sequence is the single robot connection mode, the main control module controls the front connecting arm to drive the gripper mechanism to move along the docking direction, the mechanical docking mechanisms mesh with each other to complete the mechanical docking, and detects the conduction status of the docking electrodes. If the conduction is confirmed, the connection is completed.

[0155] When the current action mode is the single robot separation mode, the main control module controls the gripper mechanism to release the mechanical docking mechanism and drives the front connecting arm to move along the separation direction to complete the separation of the single robot.

[0156] Through the above sub-steps, each execution module can be precisely controlled according to the action execution sequence, ensuring that the control logic of each action mode is clear and the execution order is reasonable, thereby improving the overall stability and task completion efficiency of the robot under multiple action mode switching.

[0157] In this embodiment, the action mode set includes the following six action modes, each corresponding to a different execution module and control method of the robot:

[0158] Wheel rolling mode: Control the wheels on the wheel leg module to output torque at the target rolling speed to enable the robot to move quickly on flat terrain; at the same time, lock the hip joint module, thigh joint module, and calf joint module of the wheel leg motion module to maintain structural rigidity.

[0159] Foot-based gait mode: The hip joint module is controlled sequentially for pitch, the thigh joint module for flexion and extension, and the lower leg joint module for posture stability, thereby achieving the foot-based gait swing and support cycle.

[0160] Gripper support mode: The gripper mechanism extends vertically and contacts the ground to form an additional support point, which is used to improve lateral or longitudinal stability and reduce center of gravity sway.

[0161] Single robot connection mode: The control of the front connecting arm drives the gripper mechanism to move along the docking direction to complete the mechanical docking, and the connection status is confirmed by the continuity detection of the docking electrodes.

[0162] Single robot separation mode: The control gripper mechanism releases the docking electrode locking structure and drives the front connecting arm to move along the separation direction to achieve single robot separation.

[0163] By defining the constraints of the above set of action modes, the execution modules and control objectives corresponding to different action modes are clearly defined, ensuring clear control logic and avoiding the coupling confusion between mode selection and control execution. An extensible action mode framework is provided to meet the multi-action mode combination requirements of robot swarm collaborative control systems in multiple scenarios. This enables the action complexity optimization method to have complete and implementable action mode set support, improving the application scope and engineering value of the method.

[0164] In this embodiment, when generating an action execution sequence, a directed graph is constructed with action patterns as graph nodes and action cost function values ​​as edge weights. A graph search algorithm is used to search for the path with the lowest cumulative cost from the initial state to the target state, and the action nodes in the path are added to the action execution sequence in sequence.

[0165] The main control module constructs the action pattern set into a set of nodes in a directed graph, let:

[0166]

[0167] in, For each candidate action mode.

[0168] In a weighted graph, any two nodes and For directed edges between two points, if a feasible transition relationship exists, the edge weight is defined as:

[0169]

[0170] in, This is the action cost function value calculated through step S4.

[0171] Path planning is performed using a graph search algorithm to find the path with the lowest cumulative cost from the current action mode state (starting node) to the target action mode state (target node).

[0172] The search is performed using Dijkstra's algorithm, represented as:

[0173]

[0174] Where Paths is the set of all feasible paths. The action execution path with the lowest cumulative cost.

[0175] Will Each node is arranged in sequence to generate an action execution sequence, which is used in step S7 to control the wheel leg motion module and the gripper mechanism to complete the propulsion control.

[0176] By using a graph search-based path planning method to generate action execution sequences, the action pattern combination problem is abstracted into a directed graph path search problem, improving the algorithm's solution efficiency. By using the action cost function value as the edge weight, multi-objective optimization of action pattern selection is achieved. Under the condition of satisfying the path access level constraint, the action execution sequence with the minimum cumulative cost is generated, reducing unnecessary actions and structure switching, and improving the robot's execution efficiency and stability under complex tasks.

[0177] In this embodiment, when filtering candidate actions in the action pattern set and excluding action patterns that do not match the path accessibility level:

[0178] If the action mode to be evaluated includes structural form switching and gripper mechanism auxiliary support, then the cumulative number of activations of the corresponding action type in the historical action sequence within the preset time window is determined; when the cumulative number of activations is greater than or equal to the set threshold, the action mode is removed from the candidate actions.

[0179] Specifically, including:

[0180] Set a fixed-length time window, denoted as For example, 60 seconds is used to count the number of times the action pattern is activated.

[0181] For the action pattern to be evaluated, when its action type belongs to the following three categories:

[0182] Structural form switching actions (such as switching between wheel rolling mode and foot gait mode).

[0183] The gripper mechanism activation action (such as the gripper execution part in gripper support mode, single robot connection mode, and single robot separation mode).

[0184] At this point, the historical action execution sequence record is invoked and statistically analyzed within the time window. This action type The total number of times it has been executed.

[0185] Set the activation threshold for the action mode ,like:

[0186]

[0187] Then the action mode Removed from the candidate action pattern set and not included in subsequent action complexity evaluation.

[0188] This historical activation count judgment mechanism can prevent frequent switching of structural form, avoid mechanical fatigue, excessive joint wear and increased energy consumption; prevent the gripper mechanism from being repeatedly activated in a short period of time, and reduce dynamic interference during execution; as an "action mode screening mechanism" in addition to the cost function, it together with the cost function evaluation forms the core of the action complexity optimization control of this invention, and improves the rationality and stability of the action execution sequence.

[0189] In this embodiment, when performing an action that includes a single robot connection mode or a single robot separation mode, the following judgment is made:

[0190] When the target action is a single robot connection mode, it is determined whether the remaining power of the target single robot to be connected is higher than the first power threshold and whether the communication link is in a stable connection state. If either condition is not met, the target single robot connection action is removed from the candidate actions.

[0191] When the target action is in module separation mode, it is determined whether there are any abnormal markers in the historical fault records of the target robot to be separated. If there are abnormal markers, the current combination state is retained and the separation action of the robot is paused.

[0192] When the robot is currently in a non-combined state and the target action mode is modular connection mode, the following judgment is performed:

[0193] The battery level is assessed by obtaining the remaining battery level of the target connection module and comparing it with a first battery level threshold. If the remaining battery level is less than the first battery level threshold, the module is determined to have insufficient battery power and does not meet the connection conditions. In this embodiment, the remaining battery level of the target connection module (0%~100%) is obtained and compared with the first battery level threshold of 20%. If the remaining battery level is less than 20%, the module is determined to have insufficient battery power and does not meet the connection conditions.

[0194] The communication status is determined by acquiring the communication link status of the target connection module. If the communication status is unstable (e.g., the signal packet loss rate exceeds a preset threshold, or the link delay is higher than a preset delay threshold), the connection condition is determined not to be met. In this embodiment, the communication link status of the target connection module is acquired. If the signal packet loss rate exceeds 5% or the link delay is higher than 100ms, the communication status is determined to be unstable, and the connection condition is not met.

[0195] If any of the above conditions are not met, the single robot connection mode will be removed from the candidate action mode set.

[0196] When the robot is currently in a combined state and the target action mode is a single robot separation mode, the following judgment is performed:

[0197] For anomaly detection, the historical fault records of the target separation module are queried. If anomaly markers (such as poor contact of connecting electrodes, abnormal internal power supply, joint jamming, etc.) are found, it is determined that separation operation is not advisable at this time. When anomaly markers are present, the current combination state is retained first, the scheduling of module separation action is suspended, and the module separation mode is removed from the candidate action mode set.

[0198] The above-described execution condition judgment process avoids performing combined connections when the module power is insufficient or the communication is unstable, thus preventing connection failure or task interruption. It also prevents separation operations when the target module has historical fault abnormality markers, ensuring overall stability and task safety. This improves the intelligence level of feasibility judgment for single robot connection and separation actions.

[0199] In this embodiment, the calculation weights of the action cost function are dynamically adjusted according to the task type and terrain complexity, wherein:

[0200] When the task type is a high-speed movement task, the weight of the unit action energy consumption calculation is less than the weight of the action execution time calculation.

[0201] When the task type is a stable obstacle crossing task, the weight value of the attitude stability change value is greater than the weight value of the unit motion energy consumption.

[0202] When the terrain complexity is detected to be higher than the preset terrain complexity threshold, the weight of the number of structural deformations is adjusted to 1.5 times the current weight value.

[0203] In this embodiment, the value of the unit action energy consumption weight is in the range of 0.1 to 1.0;

[0204] The weight value for action execution time ranges from 0.2 to 1.5.

[0205] The weight value for the number of structural deformations ranges from 0.2 to 1.2.

[0206] The weight value for attitude stability change ranges from 0.5 to 2.0;

[0207] Specifically, when the task type is a high-speed movement task, the weight value of unit action energy consumption can be set to 0.3, which is lower than the weight value of action execution time of 0.6.

[0208] When the task type is a stable obstacle crossing task, the weight value of the attitude stability change value is set to 1.5, which is greater than the weight value of the unit motion energy consumption of 0.3.

[0209] When the terrain complexity is detected to be higher than the preset terrain complexity threshold, the weight value of the number of structural deformations is increased from 0.4 to 1.5 times the current weight value (i.e., 0.6).

[0210] The weights of each parameter in the action cost function are dynamically adjusted based on the task type and terrain complexity.

[0211] The types of tasks include high-speed movement tasks, stable obstacle crossing tasks, and precise docking tasks. Different types of tasks have different requirements for execution efficiency, energy consumption, or stability.

[0212] Terrain complexity is comprehensively assessed based on terrain type, obstacle size, and path accessibility level, reflecting the degree to which terrain constrains the complexity of structural movements.

[0213] When the task type is a high-speed movement task, the weights for calculating the action cost function are configured as follows: the weight of the action execution time is higher than the weight of the energy consumption per unit action, so as to prioritize the robot's movement efficiency in the task.

[0214] When the task type is a stable obstacle crossing task, the weights for calculating the motion cost function are configured as follows: the weight of the attitude stability change value is higher than the weight of the unit motion energy consumption, so as to prioritize the stability and safety of the robot during the obstacle crossing process.

[0215] When the terrain complexity exceeds a preset threshold, the weight of the number of structural deformations is increased to 1.5 times the original weight to reduce unnecessary structural switching in complex terrain and reduce load and execution risk.

[0216] In this embodiment, the robot swarm collaborative control system can be applied to rubble search and rescue scenarios. It needs to navigate between rubble, loose ground and collapsed buildings to complete the survey of the target area in the shortest time, while ensuring robot stability and reducing energy consumption and risks caused by structural switching.

[0217] This invention achieves a comprehensive quantitative evaluation of the execution cost of various action modes by constructing an action cost function that includes unit action energy consumption, number of structural deformations, action execution time, and posture stability changes. The action cost function is used to evaluate the complexity of wheeled rolling mode, legged gait mode, gripper support mode, and single-robot connection or separation mode, simultaneously considering motion energy consumption, structural switching complexity, execution time, and stability during action mode selection. By setting structural action rejection rules in the action mode screening stage, unnecessary structural deformations, gripper activation, and auxiliary support component activation actions are reduced, lowering structural disturbances and loads during action execution. A graph search-based path planning algorithm, combined with the action cost function, generates an action execution sequence with minimum action complexity, enabling efficient propulsion control of the robot in complex terrain environments. The robot swarm collaborative control system can dynamically adjust the weight of the action cost function according to task type and terrain complexity, realizing a task-oriented action selection strategy and improving adaptability to multiple scenarios.

[0218] It should be understood that those skilled in the art can make improvements or modifications based on the above description, and all such improvements and modifications should fall within the protection scope of the appended claims.

Claims

1. A method for cooperative control of robot swarms, characterized in that, Includes the following steps: Collect structural status information, which includes the deployment status of the wheel-leg motion module, the number of connected individual robots, the activation status of the gripper mechanism, and the current posture parameters; Collect terrain information, including terrain type, obstacle size, and path accessibility level; Construct a set of action modes, which includes wheel rolling mode, leg gait mode, gripper support mode, single robot connection mode and single robot separation mode; Define an action cost function, which is used to evaluate the execution cost of an action pattern; Based on structural state information and terrain information, the complexity of actionable actions in the action pattern set is evaluated according to the action cost function. Select the action pattern or combination of action patterns with the lowest action cost function value to generate an action execution sequence; The wheel-leg motion module and gripper mechanism are controlled according to the action execution sequence; When evaluating the complexity of the remaining candidate action patterns based on the action cost function: Obtain the unit action energy consumption, number of structural deformations, action execution duration, and attitude stability change value for each candidate action pattern; The attitude stability change value is nonlinearly amplified, and the squared attitude stability change value is used as the stability score. Calculate the logarithm of the energy consumption per unit motion to the base of the natural constant e, and multiply the result by the weight of the energy consumption per unit motion calculation to generate a score for the energy consumption per unit motion. The weight is calculated by multiplying the number of structural deformations by the number of structural deformations to generate a structural deformation score. The action execution time is exponentially amplified by multiplying the exponential value of the action execution time by the action execution time calculation weight to generate an action execution time score. The cost function value of each candidate action mode is obtained by weighted summing of stability score, unit action energy consumption score, structural deformation number score and action execution time score. Compare the cost function values ​​of all candidate action patterns, select the action pattern or combination of action patterns with the smallest cost function value, and generate the action execution sequence.

2. The robot swarm cooperative control method according to claim 1, characterized in that, When evaluating the complexity of actionable actions in the action pattern set based on structural state information and terrain information, according to the action cost function: Candidate actions in the action pattern set are filtered based on the path access level, and action patterns that do not match the path access level are excluded. The complexity of the remaining candidate action patterns is evaluated based on the action cost function; Select the action pattern or combination of action patterns with the lowest action cost function value from all action patterns that meet the filtering criteria.

3. The robot swarm cooperative control method according to claim 1, characterized in that, The energy consumption per unit motion is calculated based on the integral of the product of current and voltage of the motor built into the wheel-leg motion module during the motion execution cycle; The number of structural deformations is the number of joint-driven actions during the transition from the current structural state to the target action mode. Joint drives include unfolding, contraction, and rotation, and each occurrence is recorded as 1. The execution duration of an action is calculated using the predefined trajectory execution cycle of the action pattern; The attitude stability change value is calculated by the shift in the center of gravity position and the change in attitude angle before and after the execution of the action mode.

4. The robot swarm cooperative control method according to claim 1, characterized in that, When selecting the action pattern or combination of action patterns with the lowest action cost function value to generate an action execution sequence: If the action execution sequence includes a wheel rolling mode, the control wheel leg motion module drives the wheel set to roll at a constant speed while keeping the joint in a locked state; If the action execution sequence includes a foot gait mode, the hip joint module, thigh joint module, and calf joint module of the control wheel leg motion module will complete the swing and support transition according to the preset gait timing. If the action execution sequence includes a gripper support mode, then control the gripper mechanism to extend downward and keep it in contact with the ground, serving as an additional support point for balance control; If the action execution sequence includes a single robot connection mode or a single robot separation mode, the control of the front connecting arm drives the gripper mechanism to perform mechanical docking or release operations, and detects whether the docking electrode is conductive, and detects whether the pressure reaches the preset value through the pressure sensor to determine whether the mechanical docking is successful.

5. A robot swarm cooperative control method according to claim 1 or 4, characterized in that, When generating an action execution sequence, a directed graph is constructed with action patterns as graph nodes and action cost function values ​​as edge weights. A graph search algorithm is used to search for the path with the lowest cumulative cost from the initial state to the target state, and the action nodes in the path are added to the action execution sequence in sequence.

6. The robot swarm cooperative control method according to claim 2, characterized in that, The main control module determines whether it is necessary to separate individual robots based on the task type. If it is necessary to separate individual robots, after separation, the main control module assigns corresponding target positions or sub-tasks to each individual robot based on the task objectives and environmental information. Each separated individual robot independently generates an action execution sequence adapted to its task objectives.

7. The robot swarm cooperative control method according to claim 1, characterized in that, When performing actions involving either a single-robot connection mode or a single-robot separation mode, the following judgment is performed: When the target action is a single robot connection mode, it is determined whether the remaining power of the target single robot to be connected is higher than the first power threshold and whether the communication link is in a stable connection state. If either condition is not met, the target single robot connection action is removed from the candidate actions. When the target action is in module separation mode, it is determined whether there are any abnormal markers in the historical fault records of the target robot to be separated. If there are abnormal markers, the current combination state is retained and the separation action of the robot is paused.

8. The robot swarm cooperative control method according to claim 1, characterized in that, The weights for calculating the action cost function are dynamically adjusted based on the task type and terrain complexity, where: When the task type is a high-speed movement task, the weight of the unit action energy consumption calculation is less than the weight of the action execution time calculation. When the task type is a stable obstacle crossing task, the weight value of the attitude stability change value is greater than the weight value of the unit motion energy consumption. When the terrain complexity is detected to be higher than the preset terrain complexity threshold, the weight of the number of structural deformations is adjusted to 1.5 times the current weight value.

9. A system utilizing the robot swarm cooperative control method of claim 1, characterized in that, It includes a main control module, a measurement module, and multiple interconnected individual robots. Each individual robot includes a torso module, a front connecting arm, and a wheel-leg motion module. One end of the front connecting arm is connected to the torso module via a multi-degree-of-freedom joint, and the other end is connected to a gripper mechanism. The gripper mechanism is equipped with a pressure sensor, a mechanical docking mechanism, and a docking electrode. The wheel-leg motion module includes a hip joint module, a thigh joint module, a lower leg joint module, and a wheel foot module; there are two sets of wheel-leg motion modules, located on both sides of the torso module; The wheel-leg motion module is used to execute wheel rolling mode and leg gait mode, the gripper mechanism is used to execute gripper support mode, and also to execute the connection or separation mode of individual robots, the measurement module is used to collect structural state information and terrain information, and the main control module is used to perform motion cost function calculation and generate motion execution sequence, and control one or more individual robots through motion execution sequence.

Images