4wid-4wis overdrive mobile robot motion control system
By combining the environmental perception module, behavior decision-making module, and motion control module, the control redundancy and environmental adaptability issues of the 4WID-4WIS mobile robot control system are solved, achieving stable and adaptive motion control and improving the system's autonomous decision-making capability and engineering feasibility.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- INST OF INTELLIGENT MFG TECH JITRI
- Filing Date
- 2026-04-20
- Publication Date
- 2026-07-24
AI Technical Summary
The 4WID-4WIS mobile robot control system suffers from control redundancy, large modeling errors, lack of environmental adaptability and intelligence, and is difficult to control stably in complex environments.
By employing an environmental perception module, a behavior decision-making module, and a motion control module, combined with multi-sensor fusion, deep reinforcement learning, and model predictive control, adaptive motion mode switching and stable control are achieved.
It improves the stability and adaptability of the robot system, reduces the impact of modeling errors, and enhances control performance and real-time performance in complex environments.
Smart Images
Figure CN122450127A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of robot control, specifically to a 4WID-4WIS overdrive mobile robot motion control system. Background Technology
[0002] With the rapid development of service robots, special-purpose robots, and intelligent mobile platforms, mobile robots with high mobility and strong environmental adaptability have gradually become a research hotspot. Among them, four-wheel independent drive and four-wheel independent steering (4WID-4WIS) mobile robots have superior mobility performance compared to traditional differential or Ackerman steering robots in narrow spaces, complex terrains, and dynamic environments because each wheel has independent steering and driving capabilities.
[0003] However, the 4WID-4WIS mobile robot is a typical overacting system, with its control degrees of freedom exceeding the actual constrained degrees of freedom, resulting in serious control redundancy. In practical applications, it suffers from the following main shortcomings:
[0004] (1) Most existing methods rely on accurate dynamic models or kinematic models, while the 4WID-4WIS system has a complex structure and the model parameters are difficult to calibrate accurately. In real environments, modeling errors are prone to accumulate, leading to a decline in control performance.
[0005] (2) Traditional control methods usually adopt the separation of planning and control, decoupling path planning from underlying control, lacking the ability to adapt to environmental changes in real time, and are prone to instability or collision risks in dynamic obstacles or complex scenarios.
[0006] (3) Although some rule-based or optimization-based methods can achieve multi-mode motion, mode switching depends on manually set thresholds, lacks intelligence and adaptability, and is difficult to balance system stability and motion efficiency.
[0007] (4) Existing research based on reinforcement learning often directly outputs continuous control quantities, which is difficult to train and has poor stability, making it difficult to apply directly to real overdrive robot systems. Summary of the Invention
[0008] To address the aforementioned problems, this invention proposes a 4WID-4WIS overdriven mobile robot motion control system, which integrates environmental perception and intelligent decision-making mechanisms to improve the stability, adaptability, and engineering feasibility of the robot system. The technical solution provided by this invention is as follows:
[0009] A motion control system for a 4WID-4WIS overdriven mobile robot includes an environmental perception module, a behavior decision-making module, and a motion control module.
[0010] The environment perception module is used to acquire the robot's current state information and surrounding environment information, and constructs a unified state description based on multi-sensor fusion. The multi-sensor includes at least one or more of LiDAR, depth camera, inertial measurement unit (IMU), and wheel speed odometer. The state information output by the environment perception module includes: robot pose information, linear velocity and angular velocity information, obstacle distribution information, and local or global map information.
[0011] The behavior decision module is built based on a deep reinforcement learning algorithm and includes a policy network and a value network. The behavior decision module takes the state vector output by the environment perception module as input and outputs the current optimal action mode number through the policy network. The action mode number is a discrete action space, and each number corresponds to a predefined motion control mode.
[0012] The motion control module has multiple built-in motion control modes, each corresponding to an independent motion controller. After receiving the action mode number output by the behavior decision module, the motion control module automatically switches to the motion control mode and outputs executable low-level control command signals. The control commands include: the angle control of the four steering wheels and the speed control of the four drive wheels.
[0013] Preferably, the motion control modes include free movement mode, zero-radius rotation mode, and pure translation mode.
[0014] Preferably, deep reinforcement learning algorithms use Markov decision processes to model the interaction between the agent and the environment, obtaining rewards by observing the current state and performing actions. Specifically:
[0015] The observation space contains the key information needed for decision-making, and is defined as follows:
[0016]
[0017]
[0018]
[0019]
[0020] in, The observed state at time t; To transform the reference trajectory into the robot's coordinate system, The positions of the first m reference points; One-hot encoding for actions, For the most recent n actions; The turning angle of the 4WIS-4WID robot. These are the steering angles of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively.
[0021] The action space is a set of discrete actions that can be defined for the decision-making behavior of a mobile robot, as follows:
[0022]
[0023] in Let t be the action at time t. , , These correspond to free movement mode, zero-radius rotation mode, and pure translation mode, respectively.
[0024] The reward function is defined as a weighted sum of multiple reward items:
[0025]
[0026] in, Let be the total reward at time t. , , and The weight of each reward item; A reward for moving toward the target point; The reward is the distance from the robot to the nearest point on the reference path; Reward the robot's orientation relative to the reference direction; This is a mode-switching reward system, used to penalize frequent switching and encourage the correct selection of the exercise mode.
[0027] Preferably, the system employs a simplified kinematic model:
[0028]
[0029] in, and Let be the linear velocities of the robot in the x and y directions at time t, respectively. Let t be the robot's angular velocity. Let be the vehicle's vector velocity at time t. and Let be the steering angles of the front and rear virtual wheels at time t, respectively. This represents the distance from the center of the vehicle body to the front / rear virtual wheels.
[0030] Preferably, the motion controller is constructed using a model predictive control method. The current state of the robot is used as the initial state, and the future state of the system is predicted in a rolling manner within the prediction time domain. A quadratic objective function is constructed using trajectory tracking error, control input smoothness, and energy consumption indicators. The optimal control input sequence is obtained by solving the optimization problem online. The control input includes virtual wheel rotation angle and virtual wheel drive speed, which are converted into the actual steering wheel angle and drive wheel speed of the 4WIS-4WID robot after inverse mapping. Linear or nonlinear constraints are introduced during the optimization process to limit the wheel speed range, steering angle range, and the robot's maximum angular velocity, thereby ensuring that the control commands are within the physically executable range.
[0031] Preferably, the specific process of constructing a motion controller using the model predictive control method is as follows:
[0032] definition , The formula is obtained as follows:
[0033]
[0034] Linearizing and discretizing equation (4) at any point, and ignoring the period, we get:
[0035]
[0036]
[0037]
[0038]
[0039] in, The input is controlled at time k. For the system matrix, For the control matrix, It is a 0 matrix. It is the identity matrix. and These are the reference values for the steering angles of the front and rear virtual wheels, respectively. This is the reference value for the vehicle body's vector velocity. For the observation matrix, Let m be the system state at time k, m be the dimension of control variables, and n be the dimension of state variables; the optimization problem is constructed as a quadratic programming problem, with the objective function being:
[0040]
[0041]
[0042]
[0043]
[0044]
[0045]
[0046] in, The state weight matrix is... To control the weight matrix, The length of the observation domain. The length of the control domain is defined by the following constraints:
[0047]
[0048]
[0049]
[0050]
[0051] in, For minimum control, For maximum control.
[0052] Preferably, in pure translation mode, the constraint relationship between the steering angle of each steering wheel and the speed of the drive wheel is as follows:
[0053]
[0054] in, This is a virtual steering angle. For vehicle body vector velocity, The velocities of the left front wheel, right front wheel, left rear wheel, and right rear wheel are respectively represented. The pure translational kinematic model is as follows:
[0055]
[0056] The linearized form is obtained as follows:
[0057]
[0058] The system constructs a pure translational kinematic model, establishes state update equations based on linearized form, and uses model predictive control methods to generate optimal control inputs online.
[0059] Preferably, in the zero-radius rotation mode, the constraint relationship between the wheels of the vehicle is as follows:
[0060]
[0061] in, The distance from the center of the vehicle body to the side. For the robot's angular velocity, The value is 1 when the wheel is located at the left front wheel or right rear wheel, and -1 otherwise. The system automatically generates steering angle and drive wheel speed control commands according to the preset wheel system constraint relationship, thereby realizing stable in-situ rotation control of the robot.
[0062] Compared with the prior art, the beneficial effects achieved by the present invention are:
[0063] This invention employs reinforcement learning for behavioral decision-making, avoiding precise modeling of complex overdriven systems and reducing the impact of modeling errors on control performance. The system can automatically select the optimal motion mode based on environmental changes, achieving autonomous decision-making and stable control in complex environments. By transforming complex continuous control problems into motion mode selection and intra-mode control, the system's stability and interpretability are significantly improved. Compared to methods that directly output continuous control quantities, this invention uses discrete motion mode decision-making, resulting in more stable reinforcement learning training, easier deployment in real robot systems, and flexible addition of new motion modes or sensor types without requiring a reconstruction of the overall system architecture. Attached Figure Description
[0064] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:
[0065] Figure 1 This is a kinematic schematic diagram of the 4WIS-4WID robot of the present invention. Detailed Implementation
[0066] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0067] To make the above-mentioned objectives, features and effects of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.
[0068] Example 1: A motion control system for a 4WID-4WIS overdriven mobile robot, comprising an environmental perception module, a behavior decision module, and a motion control module.
[0069] The environment perception module acquires the robot's current state information and surrounding environment information, constructing a unified state description based on multi-sensor fusion. The multi-sensor fusion includes at least one or more of LiDAR, depth camera, inertial measurement unit (IMU), and wheel speed odometer. The state information output by the environment perception module includes: robot pose information, linear and angular velocity information, obstacle distribution information, and local or global map information.
[0070] The behavior decision-making module is built based on a deep reinforcement learning algorithm and includes a policy network and a value network. The module takes the state vector output by the environment perception module as input and outputs the current optimal action mode number through the policy network. The action mode number represents a discrete action space, with each number corresponding to a predefined motion control mode. The behavior decision-making module learns the mapping relationship from the state space to the action mode space through reinforcement learning, enabling adaptive switching between different motion modes.
[0071] The motion control module incorporates multiple motion control modes, each corresponding to an independent controller function. Based on the currently selected mode, it generates wheel system control commands. These commands include: angle control values for the four steering wheels and speed control values for the four drive wheels. Upon receiving the action mode number output by the behavior decision module, the motion control module automatically switches to the corresponding control mode and outputs executable low-level control signals.
[0072] Furthermore, due to the strong coupling, time-varying nature, and multivariable nonlinearity of the 4WIS-4WID robot system, direct control based on the complete dynamic model would result in a complex control system structure, heavy computational burden, and difficulty in meeting real-time requirements. Therefore, this invention simplifies the kinematic model of the 4WIS-4WID robot through equivalent means.
[0073] like Figure 1 As shown, the four actual wheels of the 4WIS-4WID robot are equivalent to two virtual wheels, thus constructing a single-track model. In this equivalent model, the mapping relationship between the rotation angle of the virtual wheel and the linear and angular velocities of the robot body is established through Ackerman geometry, thereby achieving an approximate description of the overall motion behavior of the 4WIS-4WID robot. According to Ackerman geometry, the relationship between the control variables of the virtual wheel and the robot body velocity is as follows:
[0074]
[0075] in, and These are the velocities of the vehicle body in the x and y directions, respectively. For vehicle body vector velocity, Let ω be the angular velocity of the vehicle body. and These are the steering angles of the front and rear virtual wheels, respectively. This represents the distance from the center of the vehicle body to the front / rear virtual wheels.
[0076] By limiting the virtual wheel rotation angle to a preset range, such as within 40°, the trigonometric functions can be approximated, thereby further simplifying the kinematic model and making it meet the real-time calculation requirements of linear controllers or model predictive controllers. The trigonometric functions can be approximated to obtain:
[0077]
[0078] Substituting formula (2) into formula (1) yields a simplified kinematic model:
[0079]
[0080] in, and Let be the linear velocities of the robot in the x and y directions at time t, respectively. Let t be the robot's angular velocity. Let be the vehicle's vector velocity at time t. and These are the steering angles of the virtual wheels before and after time t, respectively. This modeling method significantly reduces the system dimensionality, effectively reduces the amount of control computation, and maintains an effective description of the robot's overall motion characteristics.
[0081] Based on the simplified kinematic model described above, this invention employs Model Predictive Control (MPC) to construct a robot motion controller. Specifically, the robot's current state is used as the initial state. Rolling predictions of the system's future state are performed within the prediction time domain. A quadratic objective function is constructed using trajectory tracking error, control input smoothness, and energy consumption indicators. The optimal control input sequence is obtained by solving the optimization problem online. The control inputs include virtual wheel rotation angles and virtual wheel drive speeds, which are converted into the actual steering wheel angles and drive wheel speeds of the 4WIS-4WID robot after inverse mapping. Simultaneously, linear or nonlinear constraints are introduced during the optimization process to limit the wheel speed range, steering angle range, and the robot's maximum angular velocity, thereby ensuring that the control commands remain within the physically executable range. Definitions , The formula is obtained as follows:
[0082]
[0083] By linearizing and discretizing equation (4) at any point, and ignoring the period, we obtain:
[0084]
[0085]
[0086]
[0087]
[0088]
[0089] in, The input is controlled at time k. For the system matrix, For the control matrix, It is a 0 matrix. It is the identity matrix. and These are the reference values for the steering angles of the front and rear virtual wheels, respectively. This is the reference value for the vehicle body's vector velocity. For the observation matrix, Let k be the system state at time k. For the kinematic model, m is the dimension of control variables and n is the dimension of state variables.
[0090] Based on the model predictive control strategy, the optimization problem is constructed as a quadratic programming (QP) problem, with the objective function being:
[0091]
[0092]
[0093]
[0094]
[0095]
[0096]
[0097] in, The state weight matrix is... To control the weight matrix, The length of the observation domain. To control the length of the domain, Let k be the generalized control increment at time k. By adding linear or nonlinear constraints, the 4WIS-4WID robot can quickly and smoothly track a given trajectory. The constraints are as follows:
[0098]
[0099]
[0100]
[0101]
[0102]
[0103] in, For minimum control, For maximum control, Controlled at time k, Let be the generalized control variable at time k. In summary, the optimization problem can be expressed as:
[0104]
[0105] This model predictive control enables robots to achieve smooth trajectory tracking in complex environments, balancing control accuracy and system stability.
[0106] To achieve autonomous decision-making and intelligent control of robots in complex environments, this invention introduces a reinforcement learning-based behavioral decision-making mechanism for adaptive switching between different motion modes. The reinforcement learning framework employs Markov decision process modeling, where the agent interacts with the environment, obtaining rewards by observing the current state and performing actions.
[0107] The observation space contains the key information needed for decision-making, and is defined as follows:
[0108]
[0109]
[0110]
[0111]
[0112] in, The observed state at time t; To transform the reference trajectory into the robot's coordinate system, The positions of the first m reference points; One-hot encoding for actions, For the most recent n actions; The turning angle of the 4WIS-4WID robot. These are the steering angles of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively.
[0113] The action space is a set of discrete actions that can be defined to represent the decision-making behavior of a mobile robot. By specifying control modules for different motion modes, stable and accurate path tracking can be achieved. The action set is as follows:
[0114]
[0115] in Let t be the action at time t. , , Corresponding to free movement mode, zero-radius rotation mode, and pure translation mode respectively, this set of actions allows the robot to switch modes according to the current state and task requirements.
[0116] The design of the reward function is a crucial step in reinforcement learning. The agent receives a reward or penalty after performing an action. Therefore, the total reward at time t is... Defined as the weighted sum of multiple reward items:
[0117]
[0118] in , , and The weight of each reward item determines its importance in the total reward; Rewards for moving towards the target point; The reward is the distance from the robot to the nearest point on the reference path; Reward the robot's orientation relative to the reference direction; This is a mode-switching reward system, used to penalize frequent switching and encourage the correct selection of the exercise mode.
[0119] Example 2: In certain application scenarios, such as narrow passages or lateral alignment tasks, the robot needs to perform pure translational motion without rotation. This example defines a pure translational motion mode. In this mode, the four steering wheels of the 4WIS-4WID robot maintain a consistent rotation angle, and the speeds of each drive wheel are coordinated and controlled according to the target translation direction, thereby enabling the robot to translate in any direction. In the pure translational mode, the constraint relationship between the rotation angle of each steering wheel and the speed of the drive wheel is as follows:
[0120]
[0121] in, This is a virtual steering angle. These represent the velocities of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively. The pure translational kinematic model is as follows:
[0122]
[0123] We can obtain:
[0124]
[0125] The system constructs a pure translation kinematic model and establishes a state update equation based on a linearized form. It then uses model predictive control methods to generate the optimal control input online, thereby achieving high-precision translation control.
[0126] Example 3: In tasks involving turning in place or adjusting attitude, the robot needs to perform zero-radius rotation. This example defines a zero-radius rotation mode. In this mode, the robot's linear velocity is zero, the steering angles of the four wheels point towards the instantaneous center of rotation according to geometric constraints, and the speeds of each drive wheel are set to positive and negative directions based on the wheel's position, enabling the robot to rotate around its own center of mass. The constraint relationships between the vehicle's wheels are as follows:
[0127]
[0128] in, The distance from the center of the vehicle body to the side. For the robot's angular velocity, The value is 1 when the wheel is located at the left front wheel or right rear wheel, and -1 otherwise. The system automatically generates steering angle and drive wheel speed control commands according to the preset wheel system constraint relationship, thereby realizing stable in-situ rotation control of the robot.
[0129] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.
Claims
1. A motion control system for a 4WID-4WIS overdrive mobile robot, characterized in that, It includes an environmental perception module, a behavior decision-making module, and a motion control module; The environment perception module is used to acquire the robot's current state information and surrounding environment information, and constructs a unified state description based on multi-sensor fusion. The multi-sensor includes at least one or more of LiDAR, depth camera, inertial measurement unit (IMU), and wheel speed odometer. The state information output by the environment perception module includes: robot pose information, linear velocity and angular velocity information, obstacle distribution information, and local or global map information. The behavior decision module is built based on a deep reinforcement learning algorithm and includes a policy network and a value network. The behavior decision module takes the state vector output by the environment perception module as input and outputs the current optimal action mode number through the policy network. The action mode number is a discrete action space, and each number corresponds to a predefined motion control mode. The motion control module has multiple built-in motion control modes, each corresponding to an independent motion controller. After receiving the action mode number output by the behavior decision module, the motion control module automatically switches to the motion control mode and outputs executable low-level control command signals. The control commands include: the angle control of the four steering wheels and the speed control of the four drive wheels.
2. The 4WID-4WIS overdrive mobile robot motion control system according to claim 1, characterized in that, The motion control modes include free movement mode, zero-radius rotation mode, and pure translation mode.
3. The 4WID-4WIS overdrive mobile robot motion control system according to claim 2, characterized in that, Deep reinforcement learning algorithms use Markov decision processes to model the interaction between an agent and its environment. The agent receives rewards by observing its current state and performing actions. Specifically: The observation space contains the key information needed for decision-making, and is defined as follows: ； ； ； ； in, The observed state at time t; To transform the reference trajectory into the robot's coordinate system, The positions of the first m reference points; One-hot encoding for actions, For the most recent n actions; The turning angle of the 4WIS-4WID robot. These are the steering angles of the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively. The action space is a set of discrete actions that can be defined for the decision-making behavior of a mobile robot, as follows: ； in Let t be the action at time t. , , These correspond to free movement mode, zero-radius rotation mode, and pure translation mode, respectively. The reward function is defined as a weighted sum of multiple reward items: ； in, Let be the total reward at time t. , , and The weight of each reward item; Rewards for moving towards the target point; The reward is the distance from the robot to the nearest point on the reference path; Reward the robot's orientation relative to the reference direction; This is a mode-switching reward system, used to penalize frequent switching and encourage the correct selection of the exercise mode.
4. The 4WID-4WIS overdrive mobile robot motion control system according to claim 2, characterized in that, The system employs a simplified kinematic model: ； in, and Let be the linear velocities of the robot in the x and y directions at time t, respectively. Let t be the robot's angular velocity. Let be the vehicle's vector velocity at time t. and Let be the steering angles of the front and rear virtual wheels at time t, respectively. This represents the distance from the center of the vehicle body to the front / rear virtual wheels.
5. The 4WID-4WIS overdrive mobile robot motion control system according to claim 4, characterized in that, The motion controller is constructed using a model predictive control method. Taking the robot's current state as the initial state, it performs rolling predictions of the system's future state within the prediction time domain. A quadratic objective function is constructed using trajectory tracking error, control input smoothness, and energy consumption indicators. The optimal control input sequence is obtained by solving the optimization problem online. The control inputs include virtual wheel rotation angles and virtual wheel drive speeds, which are converted into the actual steering wheel angles and drive wheel speeds of the 4WIS-4WID robot after inverse mapping. Linear or nonlinear constraints are introduced during the optimization process to limit the wheel speed range, steering angle range, and the robot's maximum angular velocity, thereby ensuring that the control commands are within the physically executable range.
6. The 4WID-4WIS overdrive mobile robot motion control system according to claim 5, characterized in that, The specific process of constructing a motion controller using the model predictive control method is as follows: definition , The formula is obtained as follows: ； Linearizing and discretizing equation (4) at any point, and ignoring the period, we get: ； ； ； ； in, The input is controlled at time k. For the system matrix, For the control matrix, It is a 0 matrix. It is the identity matrix. and These are the reference values for the steering angles of the front and rear virtual wheels, respectively. This is the reference value for the vehicle body's vector velocity. For the observation matrix, Let m be the system state at time k, m be the dimension of control variables, and n be the dimension of state variables; the optimization problem is constructed as a quadratic programming problem, with the objective function being: ； ； ； ； ； ； in, The state weight matrix is... To control the weight matrix, The length of the observation domain. The length of the control domain is defined by the following constraints: ； ； ； ； in, For minimum control, For maximum control.
7. The 4WID-4WIS overdrive mobile robot motion control system according to claim 6, characterized in that, In pure translation mode, the constraint relationship between the steering angle of each steering wheel and the speed of the drive wheel is as follows: ； in, This is a virtual steering angle. For vehicle body vector velocity, The velocities of the left front wheel, right front wheel, left rear wheel, and right rear wheel are respectively represented. The pure translational kinematic model is as follows: ； The linearized form is obtained as follows: ； The system constructs a pure translational kinematic model, establishes state update equations based on linearized form, and uses model predictive control methods to generate optimal control inputs online.
8. The 4WID-4WIS overdrive mobile robot motion control system according to claim 6, characterized in that, In zero-radius rotation mode, the constraint relationships between the vehicle's wheels are as follows: ； in, The distance from the center of the vehicle body to the side. For the robot's angular velocity, The value is 1 when the wheel is located at the left front wheel or right rear wheel, and -1 otherwise. The system automatically generates steering angle and drive wheel speed control commands according to the preset wheel system constraint relationship, thereby realizing stable in-situ rotation control of the robot.