Wheel-legged cargo robot driving control method

Abstract

The application discloses a driving control method of a wheel-legged cargo carrying robot, and relates to the field of robot control. The method comprises the following steps: collecting a current state vector of the wheel-legged cargo carrying robot; outputting a current action in response to the current state vector by a trained reinforcement learning intelligent agent, wherein the current action comprises a first parameter adjustment amount and a second parameter adjustment amount at a current time; outputting a first control amount and a second control amount of the wheel-legged cargo carrying robot at the current time by a first feedback controller and a second feedback controller respectively, wherein the control parameters of the first feedback controller and the second feedback controller at the current time are determined according to the first parameter adjustment amount and the second parameter adjustment amount at the current time; fusing the first control amount and the second control amount at the current time by taking a mode participation coefficient as a fusion weight to obtain a current fusion control amount; and performing real-time control on the wheel-legged cargo carrying robot according to the current fusion control amount. The method has strong robustness.

Description

Technical Field

[0001] This application relates to the field of robot control, and in particular to a drive control method for a wheeled-legged cargo robot. Background Technology

[0002] Wheel-legged cargo robots combine the high efficiency of wheeled robots with the obstacle-crossing capabilities of legged robots, making them suitable for cargo-carrying tasks requiring traversing steps, potholes, slopes, and uneven terrain, such as warehousing logistics, park transportation, and field transfers. However, the dynamic characteristics of wheeled and legged modes of wheel-legged cargo robots differ significantly, and a strong coupling exists in the hybrid wheel-leg mode, leading to complex control strategy design. Currently, most wheel-legged cargo robots use PID controllers. However, traditional PID controllers or fuzzy PID controllers require extensive manual parameter tuning, have long development cycles, and need to be retuned after robot configuration changes or task updates, resulting in poor generalization. Furthermore, traditional PID controllers or fuzzy PID controllers use fixed parameters within the same task cycle, but fluctuations in robot load and sudden terrain changes introduce external disturbances and uncertainties, causing a decline in the performance of controllers with fixed parameters, resulting in poor robustness. Summary of the Invention

[0003] The purpose of this application is to provide a drive control method for a wheeled-legged cargo robot with strong robustness.

[0004] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a drive control method for a wheeled-legged cargo robot. The wheeled-legged cargo robot includes a body for carrying goods and wheeled-leg mechanisms symmetrically distributed on both sides below the body. Each wheeled-leg mechanism includes a thigh structure, a lower leg structure, and a movable wheel. The upper end of the thigh structure is rotatably connected to the body to form a first joint, and the upper end of the lower leg structure is rotatably connected to the lower end of the thigh structure to form a second joint. The movable wheel is rotatably mounted on the lower end of the lower leg structure. The drive control method includes: The current state vector of the wheel-legged cargo robot is collected. The current state vector includes the trajectory tracking error at the current moment and its integral and derivative, the rotation speed of the moving wheel, the angle and angular velocity of the first joint and the second joint, the linear velocity and attitude angle of the body and the angular velocity, terrain feature quantity and energy feature quantity; The trained reinforcement learning agent outputs a current action in response to the current state vector, the current action including the first parameter adjustment and the second parameter adjustment at the current moment; The first feedback controller and the second feedback controller output the first control quantity and the second control quantity of the wheel-legged cargo robot at the current moment, respectively. The first control quantity is the angle control quantity of the first joint and the second joint, and the second control quantity is the rotation control quantity of the moving wheel. The control parameters of the first feedback controller and the second feedback controller at the current moment are determined according to the first parameter adjustment quantity and the second parameter adjustment quantity at the current moment, respectively. The first control quantity and the second control quantity at the current moment are fused using the mode participation coefficient as the fusion weight to obtain the current fused control quantity, and the wheel-legged cargo robot is controlled in real time according to the current fused control quantity.

[0005] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a drive control method for a wheeled-legged cargo robot. A reinforcement learning agent outputs a first parameter adjustment value and a second parameter adjustment value in response to the current state vector. These two values ​​are used to determine the control parameters of the first and second feedback controllers at the current moment, thereby enabling real-time adjustment of the control parameters of the first and second feedback controllers. This adjustment conforms to the current state vector of the wheeled-legged cargo robot, allowing for precise control of the robot. Furthermore, since the current state vector includes the trajectory tracking error and its integral and derivative, the rotational speed of the moving wheels, the angle and angular velocity of the first and second joints, the linear velocity and attitude angle of the body, as well as angular velocity, terrain features, and energy features, the first and second feedback controllers can adapt to external disturbances and uncertainties caused by robot load fluctuations and sudden terrain changes. Therefore, the drive control method for the wheeled-legged cargo robot in this embodiment has high robustness. Attached Figure Description

[0006] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0007] Figure 1 This is a schematic diagram of the wheel-legged cargo robot in the embodiments of this application; Figure 2 This is a flowchart illustrating the drive control method for a wheeled-legged cargo robot in an embodiment of this application. Detailed Implementation

[0008] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0009] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0010] The drive control method for the wheeled-legged cargo robot provided in this application is used to realize the drive control of the wheeled-legged cargo robot. It is executed by the controller of the wheeled-legged cargo robot, and the generated control quantity is executed by the actuator in the wheeled-legged cargo robot. When the actuator runs, it realizes the motion control of the wheeled-legged cargo robot.

[0011] Reference Figure 1 In one embodiment, the wheeled-legged cargo robot includes a body 1 for carrying goods and wheel-leg mechanisms symmetrically distributed on both sides of the lower part of the body. Each wheel-leg mechanism includes a thigh structure 2, a lower leg structure 3, and a movable wheel 4. The upper end of the thigh structure 2 is rotatably connected to the body 1 to form a first joint, and the upper end of the lower leg structure 3 is rotatably connected to the lower end of the thigh structure 2 to form a second joint. The movable wheel 4 is rotatably mounted on the lower end of the lower leg structure 3. Specifically, drive motors can be installed on both the first and second joints, and the two drive motors control the angles of the first and second joints respectively. Simultaneously, a drive motor is also installed on the lower end of the lower leg structure 3, which is used to control the rotation of the movable wheel 4. That is, the movable wheel 4 and the drive motors at the first and second joints are actuators.

[0012] For example, the fuselage is a frame structure. A storage compartment is located above the frame structure for storing cargo. The middle section of the frame structure is hollow, used to house the controller and battery pack. Multiple symmetrical protruding grooves are arranged below the frame structure, each containing a drive motor that drives the first joint. The output of the drive motor is connected to the upper end of the thigh structure to drive the thigh structure to rotate relative to the fuselage, thereby adjusting the angle of the first joint.

[0013] For example, the thigh mechanism adopts a hollow design, and a second motor that drives the second joint is installed inside. The output end of the second motor can be connected to the upper end of the lower leg structure through a transmission structure to drive the lower leg structure to rotate relative to the thigh structure, thereby realizing the angle adjustment of the second joint.

[0014] For example, the drive motor at the lower end of the lower leg structure can be mounted on the lower end of the lower leg structure through a shock-absorbing suspension, and the output end of the drive motor is connected to the moving wheel to drive the moving wheel to rotate.

[0015] Reference Figure 2 The drive control method for the wheeled-legged cargo robot includes steps S110 to S140.

[0016] Step S110: Collect the current state vector of the wheel-legged cargo robot. The current state vector includes the trajectory tracking error at the current moment and its integral and derivative, the rotation speed of the moving wheel, the angle and angular velocity of the first and second joints, the linear velocity and attitude angle of the body and its angular velocity, terrain features and energy features.

[0017] The drive motors at the moving wheel, the first joint, and the second joint can be servo motors. In this case, the encoder built into the servo motor can directly obtain information such as the rotational speed of the moving wheel and the angle and angular velocity of the first and second joints. If ordinary motors are used as drive motors, additional sensors need to be installed at the first joint, the second joint, and the moving wheel to obtain information such as the rotational speed of the moving wheel and the angle and angular velocity of the first and second joints.

[0018] Simultaneously, the drive control method also needs to acquire information such as trajectory tracking error and its integral and derivative, fuselage linear velocity, attitude angle and angular velocity, and terrain features. Linear velocity, attitude angle, and angular velocity can be acquired by an attitude sensor IMU mounted on the fuselage. The actual position of the fuselage can be calculated in real time using these parameters, and the trajectory tracking error can be calculated by combining this with the ideal trajectory. Furthermore, the integral and derivative of the trajectory tracking error can be obtained. For example, terrain features include terrain categories or continuous features, such as slope and vibration indices. These terrain features can be acquired by an environmental / terrain perception sensor mounted on the fuselage. For instance, the environmental / terrain perception sensor can be a binocular camera, and the terrain features can be obtained by analyzing the images captured by the binocular camera. For example, energy features include remaining battery power, real-time power of actuators, and cumulative power consumption. The real-time power and cumulative power consumption of actuators can be estimated by a current and voltage acquisition module. Specifically, the current and voltage acquisition module collects the current and voltage of each actuator (the drive motors at the moving wheels, the first joint, and the second joint). Based on the collected current and voltage data, the real-time power and cumulative power consumption of the corresponding actuators can be estimated.

[0019] In summary, the sensor module of the wheel-legged cargo robot consists of encoders or independent sensors with equivalent functions built into each actuator, attitude sensors (IMUs), environmental / terrain perception sensors, and current and voltage acquisition modules. The drive control method is executed by the controller, which is electrically connected to each sensing unit in the sensor module to acquire information such as trajectory tracking error, rotational speed of the moving wheels, angles and angular velocities of the first and second joints, linear velocity and attitude angles of the body, angular velocity, terrain features, and energy features. Simultaneously, the controller is electrically connected to each actuator to send the generated control quantities to each actuator, and each actuator executes its own control quantity to achieve coordinated control of the wheel-legged cargo robot.

[0020] It should be noted that the controller can be directly mounted on the robot body for local control. Alternatively, the controller can be located remotely, using a wireless communication module to interact with various sensing units and actuators. In this case, the wireless communication module can be mounted on the robot body, and it connects to the sensing units and actuators via local lines. Furthermore, the voltage loop module for the wheeled cargo robot can also be mounted on the robot body, while the power module supplies power to other modules (actuators, sensing units, wireless communication module, or controller).

[0021] Step S120: The trained reinforcement learning agent outputs the current action in response to the current state vector. The current action includes the first parameter adjustment amount and the second parameter adjustment amount at the current moment.

[0022] Preferably, the reinforcement learning agent can be implemented using an Actor-Critic architecture based on the Deep Deterministic Policy Gradient (DDPG) algorithm. The DDPG algorithm can be replaced by continuous action reinforcement learning algorithms such as TD3 or SAC.

[0023] Step S130: The first feedback controller and the second feedback controller output the first control quantity and the second control quantity of the wheel-legged cargo robot at the current moment, respectively. The first control quantity is the angle control quantity of the first joint and the second joint, and the second control quantity is the rotation control quantity of the moving wheel. The control parameters of the first feedback controller and the second feedback controller at the current moment are determined according to the first parameter adjustment quantity and the second parameter adjustment quantity at the current moment, respectively.

[0024] It should be noted that since the first joint and the second joint are controlled by two drive motors respectively, the first control quantity in this embodiment includes the control quantities of the two drive motors.

[0025] In either the first or second feedback controller, the control parameters at the current moment are obtained by superimposing the control parameters at the previous moment and the target adjustment amount at the current moment; in the first feedback controller, the target adjustment amount at the current moment is determined based on the first parameter adjustment amount at the current moment; in the second feedback controller, the target adjustment amount at the current moment is determined based on the second parameter adjustment amount at the current moment.

[0026] In this embodiment, the current adjustment value of the first parameter can be directly used as the current target adjustment value in the first feedback controller, and the current adjustment value of the second parameter can be directly used as the current target adjustment value in the second feedback controller. However, preferably, considering that the first and second parameter adjustment values ​​may be unreasonable, i.e., violate some constraints, a constraint verification module can be used to perform constraint verification on the first and second parameter adjustment values, correct the parameter adjustment values ​​that violate the constraints to make them meet the constraints, and then use the corrected parameter adjustment values ​​as the corresponding target adjustment values.

[0027] Specifically, if the adjustment of the first parameter at the current moment satisfies the target constraint, then the adjustment of the first parameter at the current moment is used as the target parameter adjustment in the first feedback controller at the current moment; otherwise, the adjustment of the first parameter is corrected by one of saturation, projection, safety filter or soft constraint penalty to make it satisfy the target constraint, and the corrected adjustment of the first parameter at the current moment is used as the target parameter adjustment in the first feedback controller at the current moment. And / or, if the adjustment amount of the second parameter at the current moment satisfies the target constraint, then the adjustment amount of the second parameter at the current moment is used as the target parameter adjustment amount in the second feedback controller at the current moment; otherwise, the adjustment amount of the second parameter is corrected by using one of saturation, projection or safety filter to make it satisfy the target constraint, and the corrected adjustment amount of the second parameter at the current moment is used as the target parameter adjustment amount in the second feedback controller at the current moment.

[0028] Specifically, the target constraints include non-negativity constraints and amplitude constraints of control parameters, as well as variation constraints, saturation constraints of control quantities, motion state constraints and real-time power constraints of actuators, as well as cumulative energy consumption constraints, and stability constraints of wheel-legged cargo robots.

[0029] Among these constraints, the non-negativity constraint of the control parameters ensures that the control parameters are positive; the amplitude constraint ensures that the amplitude of the control parameters is within a preset range; and the change constraint ensures that the change of the control parameters in one iteration is within a preset range. The control quantity saturation constraint ensures that the control quantity output by the feedback controller is within a preset range. The actuator's motion state constraint, real-time power constraint, and cumulative energy consumption constraint ensure that the actuator's motion state (e.g., joint angles, joint angular velocities, and joint torques at the first and second joints, and the torque of the drive motor at the moving wheels), real-time power, and cumulative energy consumption are within preset ranges, respectively. The stability constraint ensures that the wheeled cargo robot can stably support the cargo without tipping over.

[0030] It should be noted that saturation, projection, and safety filters are all relatively mature methods for correcting variables when they violate constraints; therefore, this embodiment will not describe the correction process in detail. For example, saturation correction typically involves correcting the variable in one dimension. When a variable exceeds its maximum threshold, it is corrected to that maximum threshold. Similarly, if variable A (corresponding to the first and second parameter adjustments mentioned above) causes variable B (the variable affected by variable A, corresponding to the control quantity, real-time power, and cumulative energy consumption, etc.) to exceed its maximum threshold, then variable A is corrected so that variable B exactly meets its maximum threshold.

[0031] For example, when the feedback controller is a PID controller, the control parameters include proportional K. p Integral K i and differential K d At this point, the non-negativity constraint of the control parameters is expressed as: The amplitude constraint of the control parameter is expressed as: Among them, K pmin and K pmax These are the minimum and maximum thresholds for the ratio, respectively. imin and K imax These are the minimum and maximum thresholds for integration, respectively. dmin and K dmax These are the minimum and maximum thresholds for the derivative, respectively; the constraint on the change of the control parameter is expressed as: ;in, and These represent the ratios of the current time t to the previous time t-1. and Let be the integrals at the current time t and the previous time t-1, respectively. and Let be the differentials of the current time t and the previous time t-1, respectively. , and These are the threshold values ​​for proportional, integral, and derivative changes, respectively; the control quantity saturation constraint is expressed as: ; Among them, u, u min and u max These are the control quantities and their minimum and maximum thresholds, respectively; the motion state constraints of the actuator are described as follows: the joint angle, joint angular velocity, joint torque at the first and second joints, and the torque of the drive motor at the moving wheel do not exceed the threshold; the real-time power constraints of the actuator are expressed as follows: Among them, P t P max Given the real-time power and its maximum threshold, the cumulative energy consumption constraint of the actuator is expressed as: Among them, E budget The maximum threshold for cumulative energy consumption; stability constraints can be set to the zero-point torque being located within the supporting polygon or the equivalent stability index satisfying the threshold. Here, is the convex polygon region formed by the projection of all robot support points in contact with the ground onto the ground.

[0032] The above describes the correction of the first and second parameter adjustment values ​​through target constraints. The corrected first and second parameter adjustment values ​​are used to determine the control parameters in the first and second feedback controllers, respectively. Therefore, in this embodiment, the feedback controller does not use fixed control parameters, but rather dynamic control parameters that change in real time, to better adapt to the constantly changing working state and environment of the wheeled cargo robot.

[0033] In this embodiment, the first feedback controller and the second feedback controller are respectively one of a PI controller, a PD controller, and a PID controller. The control parameters of the PI controller include proportional and integral, the control parameters of the PD controller include proportional and derivative, and the control parameters of the PID controller include proportional, integral, and derivative. A PID controller is preferred. Specifically, the PID controller is one of a PID controller with feedforward compensation, a PID controller with filtering, or a standard PID controller.

[0034] For example, the first feedback controller and the second feedback controller can be standard PID controllers. The control parameters of a standard PID controller include proportional, integral, and derivative parameters, and the control law is: ; ; ; ;

[0035] in, This represents the control output of a standard PID controller at the current time t. , and These are the target adjustment quantities as the proportion, integral, and derivative at the current time t, respectively. , and These are the current proportion, integral, and derivative, respectively. , and These represent the proportional, integral, and differential values ​​of the previous time step t-1, respectively. and They are the current time t and the historical time, respectively. The control error. The initial values ​​of the proportional, integral, and derivative axes can be tuned using the nominal operating conditions to ensure the basic stability of the system.

[0036] It should be noted that when the first feedback controller uses a standard PID controller, This is the first control variable at the current moment; when the second feedback controller uses a standard PID controller, This is the second control variable at the current moment.

[0037] Step S140: The first and second control quantities at the current moment are fused using the mode participation coefficient as the fusion weight to obtain the current fused control quantity. The wheeled and legged cargo robot is controlled in real time based on the current fused control quantity.

[0038] Specifically, the fusion formula for the fusion control quantity is: ;

[0039] in, , , and These are the fusion control quantity, mode participation coefficient, first control quantity, and second control quantity at the current time t, respectively.

[0040] When α(t) is close to 1, the wheel-legged cargo robot mainly operates in wheel mode; when α(t) is close to 0, the wheel-legged cargo robot mainly operates in leg mode; when α(t) is between 0 and 1, the wheel-legged cargo robot operates in a hybrid wheel-leg mode. To avoid the shock of mode switching, α(t) can be smoothly transitioned through low-pass filtering or a ramp function.

[0041] Here, α(t) can be set according to rules. However, this method suffers from poor generalization and robustness. Therefore, in this embodiment, it is preferable to provide the mode participation coefficient at the current moment through the reinforcement learning agent. At this time, the data provided to the reinforcement learning agent also needs to include the motion mode and mode switching flag of the wheeled cargo robot at the current moment. That is, the current state vector also includes the motion mode and mode switching flag at the current moment; the current action also includes the mode participation coefficient at the current moment.

[0042] By using reinforcement learning to generate participation coefficients for intelligent agents, it exhibits strong generalization and robustness.

[0043] As described above, target constraints need to be imposed on the parameter adjustments output by the reinforcement learning agent. Some of these constraints can be hard constraints, meaning the control parameters are corrected using hard constraints so that the corrected parameter adjustments must satisfy the hard constraints. Simultaneously, some constraints can also be soft constraints, applied to the reward function of the reinforcement learning agent during training, so that the parameter adjustments output by the agent satisfy these soft constraints as much as possible. For example, the motion state constraints, real-time power constraints, and cumulative energy consumption constraints of the actuator, as well as the stability constraints of the wheeled-legged cargo robot, can be introduced as soft constraints into the reward function.

[0044] In a specific example, the reward function of the reinforcement learning agent during training includes a trajectory error term, an energy consumption term, a smoothness term, and a constraint penalty term. In the trajectory error term, the smaller the trajectory error of the wheeled-legged cargo robot, the higher the reward value. In the energy consumption term, the lower the real-time power or cumulative energy consumption of the actuator, the higher the reward value. In the smoothness term, the smaller the rate of change of the control variable, the higher the reward value. In the constraint penalty term, a penalty value is given when the motion state constraints of the actuator or the stability constraints of the wheeled-legged cargo robot are violated.

[0045] At this point, the reward function can be expressed as: ; Where, r t Let e ​​be the reward value at the current time t. t+1 For the trajectory tracking error (trajectory error term) at the next time step t+1, u t and u t-1 These are the fused control quantities at the current time t and the previous time t-1, respectively. The difference between them constitutes a smoothing term. Since u t The square of u is positively correlated with energy consumption, thus... t The square of w is used as the energy consumption term, Penalty is the constraint penalty term, and w e w u w s and w cThese are the weights of different items.

[0046] In summary, the drive control method for the wheeled-legged cargo robot in this embodiment has the following characteristics and technical effects: 1. By using a reinforcement learning agent to output the first and second parameter adjustment values ​​of the current state vector, which are used to determine the control parameters of the first and second feedback controllers at the current moment, respectively, real-time adjustment of the control parameters of the first and second feedback controllers is achieved. This adjustment conforms to the current state vector of the wheeled-legged cargo robot, enabling precise control of the robot. Furthermore, since the current state vector includes the trajectory tracking error and its integral and derivative, the rotational speed of the moving wheels, the angle and angular velocity of the first and second joints, the linear velocity and attitude angle and angular velocity of the body, terrain features, and energy features, the first and second feedback controllers can adapt to external disturbances and uncertainties caused by robot load fluctuations and sudden terrain changes. Therefore, the drive control method for the wheeled-legged cargo robot in this embodiment has high robustness.

[0047] 2. To address the multi-mode coupling and switching impact issues of wheeled-legged cargo robots, a mode fusion coefficient is introduced. It achieves smooth switching between wheeled / legged / hybrid modes, reducing posture impact and cargo swaying. Moreover, the mode fusion coefficients can also be generated by the reinforcement learning agent in response to the current state vector. Compared to defining the mode fusion coefficients through rules, having the reinforcement learning agent output the mode fusion coefficients makes the model switching of the wheeled / legged cargo robot more consistent with the current state vector of the wheeled / legged cargo robot, providing the most suitable mode fusion coefficients in different states.

[0048] 3. A safety constraint verification module is introduced to impose non-negativity, amplitude, rate of change, actuator, and stability constraints on the parameter adjustment quantities output by the reinforcement learning agent, thereby improving engineering safety and long-term operational reliability.

[0049] 4. A parameter adjustment mechanism based on reinforcement learning is introduced to adaptively optimize PID control parameters, while constraining and verifying the amplitude of the control quantity and the operating state of the actuator. A smoothness reward term is used to suppress drastic changes in the control quantity, thereby avoiding system oscillation and response lag caused by parameter mutations or excessive control input. Therefore, it can maintain stable trajectory tracking performance under complex load and terrain change conditions, and effectively suppress overshoot and phase lag, improving robustness and adaptability.

[0050] 5. By introducing energy consumption and smoothness terms through the reward function, we can achieve coordinated optimization of trajectory, energy consumption, and smoothness, effectively reducing energy consumption and extending battery life.

[0051] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0052] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0053] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0054] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0055] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0056] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0057] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0058] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A drive control method for a wheeled-legged cargo robot, characterized in that, The wheeled-legged cargo robot includes a body for carrying goods and wheeled-leg mechanisms symmetrically distributed on both sides of the lower part of the body. Each wheeled-leg mechanism includes a thigh structure, a lower leg structure and a moving wheel. The upper end of the thigh structure is rotatably connected to the body to form a first joint, and the upper end of the lower leg structure is rotatably connected to the lower end of the thigh structure to form a second joint. The moving wheel is rotatably mounted on the lower end of the lower leg structure. The drive control method includes: The current state vector of the wheel-legged cargo robot is collected. The current state vector includes the trajectory tracking error at the current moment and its integral and derivative, the rotation speed of the moving wheel, the angle and angular velocity of the first joint and the second joint, the linear velocity and attitude angle of the body and the angular velocity, terrain feature quantity and energy feature quantity; The trained reinforcement learning agent outputs a current action in response to the current state vector, the current action including the first parameter adjustment and the second parameter adjustment at the current moment; The first feedback controller and the second feedback controller output the first control quantity and the second control quantity of the wheel-legged cargo robot at the current moment, respectively. The first control quantity is the angle control quantity of the first joint and the second joint, and the second control quantity is the rotation control quantity of the moving wheel. The control parameters of the first feedback controller and the second feedback controller at the current moment are determined according to the first parameter adjustment quantity and the second parameter adjustment quantity at the current moment, respectively. The first control quantity and the second control quantity at the current moment are fused using the mode participation coefficient as the fusion weight to obtain the current fused control quantity, and the wheel-legged cargo robot is controlled in real time according to the current fused control quantity.

2. The drive control method for the wheeled-legged cargo robot according to claim 1, characterized in that, In either the first feedback controller or the second feedback controller, the control parameters at the current moment are obtained by superimposing the control parameters at the previous moment and the target adjustment amount at the current moment; In the first feedback controller, the target adjustment amount at the current moment is determined based on the first parameter adjustment amount at the current moment; In the second feedback controller, the target adjustment amount at the current moment is determined based on the second parameter adjustment amount at the current moment.

3. The drive control method for the wheel-legged cargo robot according to claim 2, characterized in that, If the adjustment amount of the first parameter at the current moment satisfies the target constraint, then the adjustment amount of the first parameter at the current moment is used as the target parameter adjustment amount in the first feedback controller at the current moment. Otherwise, the adjustment amount of the first parameter is corrected by one of saturation, projection, safety filter or soft constraint penalty to make it meet the target constraint, and the corrected adjustment amount of the first parameter at the current time is used as the target parameter adjustment amount in the first feedback controller at the current time. And / or, if the adjustment amount of the second parameter at the current moment satisfies the target constraint, then the adjustment amount of the second parameter at the current moment is used as the target parameter adjustment amount in the second feedback controller at the current moment; Otherwise, the second parameter adjustment amount is corrected by using one of saturation, projection, or safety filters to meet the target constraint, and the corrected second parameter adjustment amount at the current moment is used as the target parameter adjustment amount at the current moment in the second feedback controller.

4. The drive control method for the wheeled-legged cargo robot according to claim 3, characterized in that, The target constraints include non-negativity constraints, amplitude constraints, and variation constraints of the control parameters, saturation constraints of the control quantity, motion state constraints and real-time power constraints of the actuator, cumulative energy consumption constraints, and stability constraints of the wheel-legged cargo robot.

5. The drive control method for the wheel-legged cargo robot according to claim 1, characterized in that, The first feedback controller and the second feedback controller are respectively one of a PI controller, a PD controller and a PID controller; The control parameters of the PI controller include proportional and integral parameters, the control parameters of the PD controller include proportional and derivative parameters, and the control parameters of the PID controller include proportional, integral, and derivative parameters.

6. The drive control method for the wheel-legged cargo robot according to claim 5, characterized in that, The PID controller is one of the following: a PID controller with feedforward compensation, a PID controller with filtering, or a standard PID controller.

7. The drive control method for the wheel-legged cargo robot according to claim 1, characterized in that, The first feedback controller and the second feedback controller are standard PID controllers. The control parameters of the standard PID controller include proportional, integral, and derivative parameters, and the control law is: ； ； ； ； in, The control quantity output by the standard PID controller at the current time t is... , and These are the target adjustment quantities as the proportion, integral, and derivative at the current time t, respectively. , and These are the current proportion, integral, and derivative, respectively. , and These represent the proportional, integral, and differential values ​​of the previous time step t-1, respectively. and They are the current time t and the historical time, respectively. Control error.

8. The drive control method for the wheeled-legged cargo robot according to claim 1, characterized in that, The fusion formula for the fusion control quantity is: ； in, , , and These are the fusion control quantity, the mode participation coefficient, the first control quantity, and the second control quantity at the current time t, respectively.

9. The drive control method for the wheel-legged cargo robot according to claim 2, characterized in that, The current state vector also includes the motion mode at the current moment and the mode switching flag; The current action also includes the mode participation coefficient at the current moment.

10. The drive control method for the wheeled-legged cargo robot according to claim 4, characterized in that, The reward function of the reinforcement learning agent during the training process includes a trajectory error term, an energy consumption term, a smoothness term, and a constraint penalty term. In the trajectory error term, the smaller the trajectory error of the wheel-legged cargo robot, the higher the reward value is given; In the energy consumption item, the lower the real-time power or cumulative energy consumption of the actuator, the higher the reward value is given; In the smoothness term, the smaller the rate of change of the control quantity, the higher the reward value is given; In the constraint penalty item, a penalty value is given when the motion state constraint of the actuator or the stability constraint of the wheel-legged cargo robot is violated.

Images