Quadruped robot cross-country control system based on laser radar

Abstract

The invention relates to the field of electronic information, in particular to a quadruped robot cross-country control system based on laser radar, and a system host comprises a sensing module used for acquiring three-dimensional point cloud data and visual information around a robot; the data processing module is used for processing and analyzing the three-dimensional point cloud data acquired by the sensing module and outputting motion compensation and registration required by the robot in the current state; the map conversion module is used for reducing the dimension of the three-dimensional point cloud map into a two-dimensional grid map; the path planning module is used for searching an optimal path in the two-dimensional grid map; and the communication module is used for coordinating data transmission and instruction sending among the modules. By fusing the data of the laser radar and the visual sensor, the adaptability to the complex cross-country environment is remarkably improved. The introduction of the visual sensor effectively makes up the sensing blind area of the laser radar under specific conditions, so that the robot can maintain stable sensing ability and path planning effect under various terrain and illumination conditions.

Description

Technical Field

[0001] The present invention relates to the field of electronic information, and in particular to a laser radar-based quadruped robot off-road control system. Background Art

[0002] As an important sensor technology, LiDAR has been widely used in the field of robot perception and navigation. LiDAR measures the distance information of the surrounding environment by emitting laser pulses and receiving reflected signals, thereby generating high-precision three-dimensional point cloud data. This data can be used to build environmental models, identify obstacles, perform path planning and autonomous navigation.

[0003] The use of LiDAR in off-road environments can also enhance the autonomy and stability of robots in unstructured environments. With the terrain information provided by LiDAR, the robot can maintain balance in complex terrain and dynamically adjust its gait to ensure smooth passage through complex terrain. In addition, the integration of LiDAR with other sensors (such as inertial measurement units (IMUs) and cameras) can further improve navigation accuracy and robustness in off-road environments.

[0004] Although the existing technology has a certain application effect in off-road path planning, it still has many shortcomings, which are mainly reflected in the following aspects:

[0005] 1. Insufficient real-time performance: In the solution of generating 3D maps by LiDAR, processing a large amount of point cloud data and performing 3D mapping requires high computing resources, which limits the real-time performance of path planning. The complex terrain in off-road environments requires robots to respond and adjust paths quickly, but existing 3D mapping methods perform poorly in this regard, which can easily lead to slow responses of robots in dynamic environments;

[0006] 2. High data processing complexity: Although 3D maps provide rich environmental information, they also increase the complexity of path planning. In particular, in path planning algorithms, a large amount of 3D data needs to be searched and optimized, which increases the calculation time. In contrast, although existing 2D map methods simplify the data processing process, they often lose key terrain information, affecting the accuracy of path planning;

[0007] 3. Poor environmental adaptability: The path planning method of a single sensor has great limitations in complex off-road environments. Whether it is a solution that relies solely on lidar or vision, it is easy to fail in special terrain or extreme environmental conditions. For example, lidar may not be able to correctly perceive obstacles with low reflectivity, and visual sensors may fail in strong light or low light conditions;

[0008] 4. Poor walking stability: Existing quadruped robots usually adopt a preset gait pattern when walking in off-road environments and lack the ability to adapt to terrain changes in real time. This causes the robot to easily lose balance or choose a poor path when encountering complex terrain, affecting walking stability. Summary of the invention

[0009] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide a quadruped robot off-road control system based on laser radar.

[0010] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0011] In a first aspect of the present invention, a laser radar-based quadruped robot off-road control system is provided, wherein the robot comprises a system host, and the system host is equipped with an externally connectable port, wherein the system host comprises:

[0012] The perception module is used to obtain 3D point cloud data and visual information around the robot;

[0013] The data processing module is used to process and analyze the three-dimensional point cloud data collected by the perception module, and output the motion compensation and registration required by the current state of the robot;

[0014] A map conversion module is used to reduce the dimensionality of a three-dimensional point cloud map to a two-dimensional grid map;

[0015] Path planning module, used to find the optimal path in a two-dimensional grid map;

[0016] The communication module is responsible for coordinating data transmission and instruction sending between modules.

[0017] Preferably, the perception module includes:

[0018] The laser radar unit is used for the robot to obtain all-round three-dimensional point cloud data of the surrounding environment in real time. By emitting laser pulses and receiving reflected signals, it collects depth information of the environment and generates a high-precision three-dimensional point cloud.

[0019] The visual recognition unit is used by the robot to identify the feasibility and potential obstacles of the road ahead, and pass the obstacle information to the path planning module to adjust the path planning results in real time.

[0020] Preferably, the map conversion module slices the three-dimensional point cloud data and converts it into a two-dimensional grid map suitable for the A* algorithm, and each grid in the map represents a passable or impassable area;

[0021] The path planning module is equipped with an A* algorithm. According to the current robot position and the target position, the A* algorithm gradually expands the search area, eventually generates an optimal path from the starting point to the target point, and outputs a path point sequence.

[0022] Preferably, the data processing module runs on the system host based on the Fast-L IO algorithm, and includes the following running steps:

[0023] S1.I MU data preprocessing, filtering the data collected by the perception module to improve the accuracy of attitude estimation;

[0024] S2. Point cloud preprocessing and registration: Use the preliminary pose information provided by the IMU to perform preliminary registration on the point cloud data collected by the lidar; then use motion compensation and feature matching to ensure that the point cloud data of different frames can be consistent in space and time;

[0025] S3. Robot state estimation: Based on the information obtained in step S2, the robot's possible posture in the future is predicted based on the current robot's posture, and motion control compensation is output based on the measurement value of the IMU.

[0026] Preferably, the path planning module includes a gait control unit, which is used to plan the foot trajectory according to the information obtained by the data processing module and the perception module, and pass through complex terrain based on a suitable gait pattern.

[0027] In a second aspect of the present invention, a computer device is provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the system function of the first aspect is implemented when the processor executes the computer program.

[0028] In a third aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the system function of the first aspect is realized.

[0029] Compared with the prior art, the present invention has the following beneficial effects:

[0030] 1: This invention uses Fast-L IO odometer for real-time mapping and reduces the dimensionality of 3D point cloud map to 2D grid map, which greatly reduces the complexity and computation of data processing. This optimization not only speeds up the response speed of path planning, but also ensures that the robot can adjust the path in real time in a dynamic off-road environment, thereby improving the real-time performance and reliability of the overall system;

[0031] 2: The present invention uses MI D-360 omnidirectional 3D laser radar for all-round environmental perception, and combines IMU data to achieve high-precision pose estimation and 3D mapping. By reducing the dimensionality of 3D point cloud data to a 2D map, and combining the road and obstacle recognition capabilities of the visual recognition module, the system can fully and accurately perceive the terrain features and potential obstacles in complex off-road environments, thereby ensuring the accuracy of path planning;

[0032] 3: Compared with the limitations of the single sensor path planning solution in the prior art, this invention significantly improves the adaptability to complex off-road environments by integrating the data of the laser radar and the visual sensor. The introduction of the visual sensor effectively compensates for the blind spot of the laser radar under certain conditions, enabling the robot to maintain stable perception and path planning effects under a variety of terrain and lighting conditions;

[0033] 4: The present invention uses a gait control module to dynamically adjust the gait mode of the quadruped robot according to different terrain and path conditions, ensuring that the robot can smoothly and safely pass through complex terrain. This gait optimization strategy not only improves the robot's movement efficiency, but also reduces the frequency of posture adjustment caused by terrain changes, thereby extending the robot's endurance and further enhancing its ability to perform off-road tasks. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0035] Figure 1 is a flow chart of the visual recognition module of the present invention;

[0036] Figure 2 is a flow chart of a data processing module of the present invention;

[0037] Figure 3 is a flow chart of the path planning module of the present invention;

[0038] Figure 4 It is a flow chart of the gait control unit of the present invention. DETAILED DESCRIPTION

[0039] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0040] Example 1

[0041] like Figure 1-2As shown, the present invention provides a laser radar-based quadruped robot off-road control system, which includes a perception module, a data processing module, a map conversion module, a path planning module and a communication module on the system host of the quadruped robot, wherein:

[0042] The perception module includes:

[0043] The laser radar unit uses the MID-360 omnidirectional 3D laser radar installed in front of the quadruped robot. It emits laser pulses and receives reflected signals to collect depth information of the environment to generate a high-precision three-dimensional point cloud. The laser radar continuously scans the surrounding environment at a predetermined frequency, and the collected point cloud data is transmitted to the data processing module in real time;

[0044] The visual recognition unit uses a visual sensor installed on the front of the robot, such as Figure 1 As shown, the visual recognition module runs on the host, located in the camera thread, and is used to identify the feasibility and potential obstacles of the road ahead. The yolov10 target detection model is used. The system marks the identified obstacles and passes the obstacle information (depth, position, etc.) to the path planning module to adjust the path planning results in real time.

[0045] The data processing module, based on the Fast-LI0 algorithm running on the host, integrates the IMU (inertial measurement unit) and the point cloud data from the MID-360 lidar to estimate the robot's position and posture in real time and generate a three-dimensional point cloud map of the environment. The specific steps are as follows:

[0046] Step 1: IMU data preprocessing: In the Fast-LI0 system, IMU data preprocessing is crucial to ensure that noise and drift are removed to improve the accuracy of attitude estimation. IMU data usually includes acceleration (a IMU ) and angular velocity (ω IMU ), these data will affect the attitude estimation of the system. We can use the integral formula to calculate the attitude. First, the angular velocity data of the IMU is expressed as an antisymmetric matrix:

[0047]

[0048] Among them, R(t) represents the rotation matrix of the posture, The antisymmetric matrix representing the angular velocity, t1 is the end time, t0 is the initial time, and t′ is the integral; in this process, the noise can be effectively removed by filtering the data (such as Kalman filtering).

[0049] Step 2: In point cloud preprocessing, the initial pose information provided by the IMU is first used to perform preliminary registration on the point cloud data collected by the LiDAR. Assume that the point cloud data of the LiDAR is Plidar , through the attitude information of IMU, the point cloud data can be converted into a relative coordinate system using the following formula:

[0050] P aligned =R IMU P lidar +t IMU

[0051] Among them, R IMU represents the rotation matrix estimated by IMU, t IMU is the translation vector estimated by the IMU.

[0052] Step 3: Point cloud preprocessing and registration. In point cloud preprocessing, the preliminary pose information provided by IMU is first used to perform preliminary registration on the point cloud data collected by the lidar. Assume that the point cloud data of the lidar is P lidar , through the attitude information of IMU, the point cloud data can be converted into a relative coordinate system using the following formula:

[0053] P aligned =R IMU P lidar +t IMU

[0054] Among them, R IMU represents the rotation matrix estimated by IMU, t IMU is the translation vector estimated by the IMU.

[0055] In Fast-LI0, point cloud registration is performed by motion compensation and feature matching. The purpose of registration is to ensure that point cloud data of different frames can maintain consistency in space and time. The method used is usually ICP (iterative closest point algorithm) or NDT (normal distribution transform). The goal of point cloud registration is to find the relative transformation T between two frames of point cloud data. k , which is achieved by minimizing the following objective function:

[0056]

[0057] Among them, P k，i and P k-1，i Represent the point cloud data of the kth frame and the k-1th frame respectively; T is T representing the minimum value k .

[0058] Step 4: Robot state estimation. The state estimation in the Fast-LI0 system combines the high-frequency data of the IMU and the precise perception capability of the lidar, and uses the extended Kalman filter (EKF) to estimate the robot's position and posture. The state vector x of the system t Contains the robot's position, velocity and posture. The prediction equation is as follows:

[0059] x t|t-1 =f(x t-1 ,u t )

[0060] where u t is the control input, which represents the measurement value of IMU, and t is the frame number. The measurement equation is as follows:

[0061] x t|t =x t|t-1 +K t (z t -h(x t|t-1 ))

[0062] Among them, K t is the Kalman gain, z t is the observation data of the lidar, and h(·) is the observation model.

[0063] Step 5: Map update. In Fast-LI0, the map is composed of a 3D point cloud generated by the LiDAR. As the robot moves, the map is continuously updated. Assume that the current map is M t , the last updated map was M t-1 , the newly observed point cloud is P t , then the map update formula is as follows:

[0064] M t =M t-1 ∪P t

[0065] In order to improve the processing efficiency of the map, the map data can be sparsely processed by using the grid occupation or voxelization method. The overall process is as follows: Figure 2 shown.

[0066] Map conversion module: The system selects the appropriate height to slice the point cloud data according to the robot's travel requirements, so that the map conversion module slices the three-dimensional point cloud data and converts it into a two-dimensional grid map suitable for the A* algorithm. Each grid in the map represents a passable or impassable area.

[0067] Path planning module, such as Figure 3 As shown in the figure, according to the current robot position and the target position, the A* algorithm gradually expands the search area. The A* algorithm evaluates the cost function of each node and finally generates an optimal path from the starting point to the target point, outputs a sequence of path points, and gives the quadruped robot a yaw angle θ and a terrain slope β.

[0068] In the path planning module, the robot will continue to move forward during the planning process, causing the perception module to continuously collect information about the surrounding environment. The collection of environmental information will affect the robot's gait pattern. Therefore, the path planning module and the perception module will eventually affect each other; the gait control unit included in the path planning module includes an algorithm that combines position control and force control to select an appropriate gait according to terrain conditions, such as Figure 4 As shown, specifically:

[0069] 1) Gait function: The gait function of a quadruped robot is a mathematical model used to describe the movement of the robot's legs when walking or moving. It determines how the robot maintains balance and moves forward by moving its legs alternately. Through bionics analysis and multiple experiments, the work uses Bezier curves to plan the foot trajectory and design a reasonable gait function;

[0070] 2) VMC model:

[0071] The VMC (Virtual Model Control) model is an intuitive control strategy that generates virtual forces by introducing virtual components (such as springs, dampers, bearings, etc.) at the robot's internal action points or between it and the external environment, thereby guiding the robot to perform the desired action. These virtual forces are converted into the required joint torques through the calculation of the Jacobian matrix, and then used as input signals for joint control to cause the robot to move to achieve effects similar to those of the virtual components. In order to achieve the conversion of forces or torques from the task space to the joint space, the relationship between the two spaces is first determined, which involves solving the forward kinematics model:

[0072] q=f(m)

[0073] In the formula, q=[q1 q2 q3...q n ] T represents the pose vector of the robot body coordinate system with respect to the ground coordinate system with n degrees of freedom, and m = [m1 m2 m3 ... m k ] T represents the position vector of k joint variables, and f represents the matrix transpose;

[0074] Calculate the partial derivative matrix of q with respect to :

[0075] δq=J·δmm

[0076] According to the principle of virtual work, we can get:

[0077] τ T δm+(-F) T δq=0

[0078] Where τ=[τ1 τ2 τ3 ... τ k ] Tis the column vector of joint torque, F=[F1 F2 F3... F n ] T is the external force. From the above formula, we can get:

[0079] τ=J T ·F

[0080] In the motion control of quadruped robots, there are some commonly used virtual components, such as springs and dampers. The role of these virtual components is to provide the necessary damping and elasticity during the robot's motion, so that the robot's motion is more stable and natural. The constitutive equations of these virtual components are:

[0081] F v =A v (s d -s)+B v (v d -v)

[0082] In the formula, F v is the virtual force generated by the virtual component, A v and B v are the elastic and damping coefficient matrices, s and v are the real-time displacement and velocity of generalized coordinates, and the subscript d represents the expected displacement and expected velocity, which are related to the gait function. The virtual force is applied to the center of mass of the robot body to control the robot body to achieve the expected motion. According to the above formula, we can get:

[0083] τ d =J T ·F v

[0084] τ d =C v (q d -q)+D v (v d -v)

[0085] Where C v , D v is the available coefficient function matrix, from which the expected moment τ is obtained d Relationship with the real-time displacement and velocity of the robot.

[0086] 3) Combined with visual module:

[0087] The A* algorithm path planning module and vision module can provide the quadruped robot with information such as the yaw angle θ and the terrain slope β. By combining these data, the formula is modified to obtain:

[0088] τ d =C v q+D v ·v+Ev ·θ+F v β+G v

[0089] Where C v , D v , E v , F v , G v is the coefficient function matrix related to the desired displacement and the desired speed. Thus, stable control of the robot movement is obtained. If a specific point of the robot (such as the center of mass height) needs to be maintained at a predetermined position, both spring and damping elements can be used for control; if the goal is to achieve a specific speed of the robot (such as horizontal movement speed), only the damping element (C v =0);

[0090] The communication module transmits the data of each sensor, path planning results and gait control instructions between modules, realizing real-time data interaction and synchronous control between modules.

[0091] Specifically, the method according to the present invention is used in an actual robot, and a conventional robot structure is adopted, including the following hardware structure: the fuselage frame is constructed using 6061 aluminum square tube, 7075 aluminum alloy, glass fiber board and PLA material, and the structure of each mechanism is optimized using topology optimization and finite element analysis to ensure strength and lightness. The fuselage size is 60 cm long, 40 cm wide and 50 cm high, and the weight of the whole machine is less than 15 kg. It uses a three-degree-of-freedom series leg configuration and has strong terrain adaptability;

[0092] A mid360 laser radar is installed at a higher position in front of the fuselage, with an IMU sensor integrated inside, which is used for 360-degree environmental scanning to generate environmental point cloud information; a D435i depth camera is installed below the laser radar to identify the feasibility and potential obstacles of the road ahead; the IMU sensor installed inside the fuselage is also used to monitor the tilt and rotation status of the robot;

[0093] An Intel mini host is installed inside the fuselage and runs the ROS operating system to process radar and depth camera data and execute path planning algorithms. The STM32 microcontroller control board communicates with the host through the serial port, is responsible for parsing the control commands sent by the host, and runs the VMC virtual model algorithm to control the coordinated movement of each motor to achieve stable movement of the machine.

[0094] Each leg is installed with three Xiaomi Gybergear motors to drive the hip joint, thigh joint and knee joint respectively. The motors are connected to the STM32 microcontroller through the CAN bus to achieve high-speed data transmission and precise control.

[0095] After the installation is completed, the method of the present invention can be used to work. After power-on, the STM32 microcontroller first performs a self-check to ensure that all hardware components are working properly; sends initialization commands to all motors to set initial position and speed parameters; starts the IMU sensor and initializes the attitude estimation algorithm; the host sends a command through the ROS system to start the lidar module and the visual perception module.

[0096] The laser radar scans the surrounding environment at a frequency of 10Hz, acquiring the 3D point cloud data of the surrounding environment in real time. After each laser radar scan, the Fast-LIO algorithm performs motion compensation and registration on the point cloud based on the IMU data, outputs the precise position of the robot's current position, and generates a real-time updated 3D map; the depth camera captures the front environment at a frame rate of 30fps, identifies obstacle information, and marks the obstacles according to their distance from the fuselage and transmits them to the path planning module; the onboard IMU sensor monitors the robot's posture changes in real time.

[0097] The 2D map generation module combines 3D point cloud data and obstacle information to generate a 2D grid map, uses the A* algorithm to plan the optimal path on the grid map, outputs a path point sequence, and sends it to the STM32 microcontroller. The STM32 microcontroller receives the path points and speed instructions sent by the host, generates the corresponding gait curve, calculates the target torque of each joint motor, sends control signals to each motor through the CAN bus, and executes the motion command.

[0098] The visual module monitors the relative position and posture of the robot and the environment in real time, and continuously identifies obstacles ahead, and transmits them to the path planning module to optimize the robot's target path. The IMU sensor feeds back the robot's real-time posture information, and the STM32 microcontroller dynamically adjusts the motor torque through the VMC algorithm based on the feedback information to adapt to terrain changes and maintain balance.

[0099] In summary, based on the existing technology, the present invention significantly improves the capabilities of the quadruped robot in path planning, environmental perception, walking stability, etc. in complex off-road environments, and has obvious technological advancement and practical application value.

[0100] Finally, it should be noted that the above description is only 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 aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A laser radar-based quadruped robot off-road control system, the robot includes a system host, the system host is equipped with an external port, characterized in that: The system host includes: The perception module is used to obtain 3D point cloud data and visual information around the robot; The data processing module is used to process and analyze the three-dimensional point cloud data collected by the perception module, and output the motion compensation and registration required by the current state of the robot; A map conversion module is used to reduce the dimensionality of a three-dimensional point cloud map to a two-dimensional grid map; Path planning module, used to find the optimal path in a two-dimensional grid map; The communication module is responsible for coordinating data transmission and instruction sending between modules.

2. According to the laser radar-based quadruped robot off-road control system of claim 1, it is characterized in that: The perception module and the path planning module are coupled to each other, and the perception module includes: The laser radar unit is used for the robot to obtain all-round three-dimensional point cloud data of the surrounding environment in real time. By emitting laser pulses and receiving reflected signals, it collects depth information of the environment and generates a high-precision three-dimensional point cloud. The visual recognition unit is used by the robot to identify the feasibility and potential obstacles of the road ahead, and pass the obstacle information to the path planning module to adjust the path planning results in real time.

3. The laser radar-based quadruped robot off-road control system according to claim 1, characterized in that: The map conversion module slices the three-dimensional point cloud data and converts it into a two-dimensional grid map suitable for the A* algorithm, where each grid in the map represents a passable or impassable area; The path planning module is equipped with an A* algorithm. According to the current robot position and the target position, the A* algorithm gradually expands the search area, eventually generates an optimal path from the starting point to the target point, and outputs a path point sequence.

4. A laser radar-based quadruped robot off-road control system according to claim 1 or 2, characterized in that: The data processing module runs on the system host based on the Fast-LIO algorithm and includes the following operating steps: S1.IMU data preprocessing: filtering the data collected by the perception module to improve the accuracy of attitude estimation; S2. Point cloud preprocessing and registration: Use the preliminary pose information provided by the IMU to perform preliminary registration on the point cloud data collected by the lidar; then use motion compensation and feature matching to ensure that the point cloud data of different frames can be consistent in space and time; S3. Robot state estimation: Based on the information obtained in step S2, the robot's current posture is used to predict possible future postures, and motion control compensation is output based on the IMU's measurement values.

5. A laser radar-based quadruped robot off-road control system according to claim 1 or 4, characterized in that: The path planning module includes a gait control unit, which is used to plan the foot trajectory according to the information obtained by the data processing module and the perception module, and pass through complex terrain based on a suitable gait pattern.

6. The laser radar-based quadruped robot off-road control system according to claim 5, characterized in that: The gait control unit includes a gait function and a VMC model, which are used to combine with the perception module to determine the state that is most suitable for the current robot.

7. A computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the system function of any one of claims 1 to 5 when executing the computer program.

8. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the system function described in any one of claims 1 to 5.

Images