A wheel-legged hybrid harvesting robot and its control method

Abstract

This invention relates to the field of agricultural automation technology, providing a wheel-legged hybrid harvesting robot and its control method. The robot includes a wheel-legged dual-mode mobile chassis, a sensing system, and a working module. The wheel-legged dual-mode mobile chassis adopts a four-legged biomimetic layout, with each leg encased in a foot-end shell. Wheels are located at the bottom of the foot-end shells, and wheel hub drive motors are housed inside the foot-end shells. The sensing system includes a lidar, a depth camera, and a control unit. The lidar is mounted at the front end of the wheel-legged dual-mode mobile chassis, and the depth camera is mounted at the execution end of the working module. The working module is mounted on the wheel-legged dual-mode mobile chassis. Through the innovative wheel-legged hybrid structure and hierarchical collaborative intelligent control strategy, the harvesting robot achieves full-process automation from efficient movement to stable operation in unstructured agricultural environments, possessing extremely high application value and promising prospects for widespread adoption.

Description

Technical Field

[0001] This invention belongs to the field of agricultural automation technology, and in particular relates to a wheel-legged composite harvesting robot and its control method. Background Technology

[0002] With the increasing demand for automation and intelligence in modern agriculture, agricultural robots, especially harvesting robots, have become a research hotspot. However, agricultural operating environments such as orchards and mountainous areas are often unstructured, with features such as uneven terrain, varying slopes, and the presence of gullies or obstacles.

[0003] Existing agricultural robots are mainly divided into three categories: wheeled, tracked, and legged. Wheeled robots have high movement speed and energy efficiency, but poor adaptability in rugged terrain, and are prone to slipping and tipping over. Tracked robots have improved obstacle-crossing ability, but are inconvenient to turn and cause greater damage to the ground. Legged robots (such as quadruped robots) have excellent terrain adaptability and stability, and can cross obstacles, but their movement speed is slow, energy consumption is high, and they are not suitable for large-scale, high-efficiency relocation operations.

[0004] Therefore, how to design a harvesting platform that can move quickly on flat surfaces like a wheeled robot and walk stably on complex terrain like a legged robot, and achieve efficient and stable harvesting in all terrains, is a technical problem that urgently needs to be solved in the field. Summary of the Invention

[0005] The purpose of this invention is to provide a wheel-legged composite harvesting robot and its control method, aiming to solve the problems mentioned in the background art.

[0006] The present invention is implemented as follows: a wheel-leg hybrid harvesting robot includes a wheel-leg dual-mode mobile chassis, a sensing system, and a working module;

[0007] The dual-mode wheel-foot mobile chassis adopts a four-legged bionic layout. Each foot is wrapped by a foot end shell. The bottom of the foot end shell is equipped with a wheel, and the inside of the foot end shell is equipped with a hub drive motor for driving the wheel to rotate. The hub drive motor has a servo locking function. When it is necessary to switch to the foot mode, its controller can make the hub drive motor enter a high-damping or locked state, thereby fixing the wheel and making the foot end a stable support point.

[0008] The perception system includes a lidar, a depth camera, and a control unit. The lidar is mounted at the front end of the wheel-legged dual-mode mobile chassis for long-range environmental scanning and SLAM mapping. The depth camera is mounted at the execution end of the working module for close-range fruit recognition, localization, and fine navigation. The control unit is located inside the wheel-legged dual-mode mobile chassis and is used to process data from various sensors and execute decision-making and control algorithms. The control unit has a built-in inertial measurement unit (IMU) for acquiring the robot's pitch and roll attitude information in real time.

[0009] The working module is mounted on a dual-mode wheel-foot mobile chassis and is used for harvesting.

[0010] In a further technical solution, the working module includes a six-axis robotic arm, which is mounted on a wheel-foot dual-mode mobile chassis. The end of the six-axis robotic arm is equipped with a gripper end effector for grasping fruit, and the depth camera is also mounted on the end of the six-axis robotic arm.

[0011] A further technical solution is that the wheel-foot dual-mode mobile chassis is equipped with a scissor lifting mechanism, and the six-axis robotic arm is mounted on the wheel-foot dual-mode mobile chassis through the scissor lifting mechanism.

[0012] A further technical solution is provided, wherein the predictive balance control algorithm steps of the wheel-foot hybrid harvesting robot are as follows:

[0013] First, the terrain data processing module in the control unit fuses the data from the lidar and IMU;

[0014] The processed data is divided into two paths: one path performs ground unevenness analysis to assess the flatness of the area to be stepped on; the other path performs balance state assessment to determine the current tilt and stability of the vehicle body.

[0015] Both sets of information are fed into the four-leg adjustment algorithm, which is based on a predictive control model and calculates the amount of joint adjustment required for each of the four legs to maintain or restore balance.

[0016] Finally, the joint control calculation module in the control unit converts these adjustments into specific control commands for the servo motors of each leg, driving the robot to complete the posture adjustment.

[0017] The entire process forms a closed loop. The actual posture of the wheel-foot dual-mode moving chassis is fed back to the control unit through the IMU for continuous and stable balance control.

[0018] A further technical solution involves the following steps for 2.5D terrain mapping of the wheel-footed hybrid harvesting robot:

[0019] First, the raw point cloud data collected by the lidar is preprocessed and the coordinates are unified. Then, the ground is segmented using an algorithm to extract the pure terrain surface point cloud.

[0020] Subsequently, the control unit performs gridding and elevation interpolation in parallel, and then merges the two to generate a DEM (Digital Elevation Model).

[0021] Finally, by analyzing the DEM, key terrain features including slope and ruggedness are extracted to form a 2.5D terrain map; the ruggedness can be quantified using indicators such as the Terrain Ruggedness Index (TRI).

[0022] A further technical solution is provided, wherein the multi-mode motion control steps of the wheel-leg hybrid harvesting robot are as follows:

[0023] The motion mode decision module in the control unit receives the calculated TRI value, slope information, and real-time obstacle detection results from the 2.5D terrain map;

[0024] When the robot senses that the terrain ahead is flat (TRI < 0.01) and there are no obstacles, it switches to wheeled high-speed mode. In this mode, the hub drive motor drives the wheels to move the robot while the leg joints remain in a fixed posture.

[0025] When a slightly uneven terrain is detected, i.e. 0.01≤TRI<0.05, the robot switches to wheeled + active suspension mode. In this mode, the robot is still driven by wheels, but the leg joints will extend and retract in a small range and at a high frequency according to the predictive balance control algorithm to absorb the impact of the ground and play the role of active suspension.

[0026] When it senses rugged terrain or needs to cross obstacles (TRI≥0.05), it switches to a slow-speed legged mode. In this mode, the hub drive motors lock to fix the wheels, and the robot traverses complex areas by alternately lifting and moving its four legs.

[0027] When the robot navigates to the designated picking point, it enters leveling mode, extends its four legs for support, and precisely adjusts the leg length to adjust the upper platform to a completely level position, providing a zero-tilt reference for the robotic arm's operation.

[0028] In any mode, if the sensors detect an emergency obstacle, the emergency obstacle avoidance mode will be triggered, and the robot will immediately stop or perform a detour.

[0029] A further technical solution is provided, wherein the collaborative precision positioning steps of the wheel-foot hybrid harvesting robot are as follows:

[0030] The SLAM module of the LiDAR is responsible for building a global map over a large area and providing pose estimation; at the same time, the visual odometry (VO) module of the depth camera provides relative motion estimation at close range.

[0031] Attitude data from the IMU, LiDAR pose, and visual odometry pose are all input into the extended Kalman filter (EKF) fusion module in the control unit;

[0032] The EKF fusion module combines data from various sensors, suppresses noise and drift, and finally outputs a 6-DOF pose, which is used by a six-axis robotic arm for coordinate transformation and precise positioning of the harvesting target.

[0033] Beneficial effects:

[0034] The present invention provides a wheel-legged composite harvesting robot and its control method, the beneficial effects of which are as follows:

[0035] (1) High terrain adaptability and high efficiency: By combining the wheel-foot composite structure with multi-mode motion control strategy, the robot can move at high speed on flat ground and switch to stable walking on foot on rugged ground, thus taking into account both movement efficiency and terrain adaptability.

[0036] (2) Strong active balancing and operational stability: Through predictive balancing control algorithms, the robot can perceive terrain undulations in real time and adjust its posture to ensure the platform remains stable during dynamic movement. At the picking point, the platform leveling function provides a stable operating benchmark for the robotic arm, significantly improving the picking success rate.

[0037] (3) Precise positioning and reliable navigation: By integrating the global mapping capability of lidar and the local fine perception capability of depth camera, high-precision and robust positioning is achieved in orchard environments with weak or missing GPS signals, which provides a guarantee for the precise operation of robotic arm.

[0038] (4) The algorithm is complete and implementable: It provides a detailed calculation process and parameter calibration method to ensure that those skilled in the art can implement the present invention based on the description, thus solving the problem of insufficient algorithm disclosure. Attached Figure Description

[0039] Figure 1 This is a schematic diagram of the structure of a wheel-legged composite harvesting robot provided in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram of the foot end of a wheel-foot hybrid harvesting robot provided in an embodiment of the present invention;

[0041] Figure 3 Block diagram of the predictive balance control algorithm;

[0042] Figure 4 Flowchart for 2.5D terrain mapping;

[0043] Figure 5 This is a state transition diagram for multi-mode motion control;

[0044] Figure 6 This is a flowchart of the collaborative localization process using lidar and depth cameras.

[0045] In the attached diagram: 1-Dual-mode mobile chassis with wheels and feet; 2-Control unit; 3-Scissor lifting mechanism; 4-Six-axis robotic arm; 5-Depth camera; 6-Grip end effector; 7-LiDAR; 8-Foot end housing; 9-Wheel hub drive motor; 10-Wheel. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0047] The specific implementation of the present invention will be described in detail below with reference to specific embodiments.

[0048] like Figure 1 and Figure 2 As shown, an embodiment of the present invention provides a wheel-leg hybrid harvesting robot, including a wheel-leg dual-mode mobile chassis 1, a sensing system, and a working module;

[0049] The dual-mode mobile chassis 1 adopts a four-legged bionic layout, with each leg end wrapped by a leg end shell 8. The bottom of the leg end shell 8 is provided with a wheel 10, and the interior of the leg end shell 8 is provided with a hub drive motor 9 for driving the wheel 10 to rotate. The hub drive motor 9 has a servo locking function. When it is necessary to switch to the leg mode, its controller can make the hub drive motor 9 enter a high-damping or locked state, thereby fixing the wheel 10 and making the leg end a stable support point.

[0050] The perception system includes a lidar 7, a depth camera 5, and a control unit 2. The lidar 7 is mounted on the front end of the wheel-legged dual-mode mobile chassis 1 for long-range environmental scanning and SLAM mapping. The depth camera 5 is mounted on the execution end of the working module for close-range fruit recognition, localization, and fine navigation. The control unit 2 is located inside the wheel-legged dual-mode mobile chassis 1 and is used to process data from various sensors (i.e., lidar 7, depth camera 5, and IMU) and execute decision-making and control algorithms. The control unit 2 has a built-in IMU for acquiring the robot's pitch and roll attitude information in real time.

[0051] The working module is installed on the wheel-foot dual-mode mobile chassis 1 and is used for harvesting.

[0052] like Figure 1 As shown, in a preferred embodiment of the present invention, the working module includes a six-axis robotic arm 4, which is mounted on a wheel-foot dual-mode mobile chassis 1. The end of the six-axis robotic arm 4 is equipped with a gripper end effector 6 for grasping fruit, and the depth camera 5 is also mounted on the end of the six-axis robotic arm 4.

[0053] like Figure 1 As shown, in a preferred embodiment of the present invention, the wheel-foot dual-mode mobile chassis 1 is provided with a scissor lifting mechanism 3, and the six-axis robotic arm 4 is mounted on the wheel-foot dual-mode mobile chassis 1 through the scissor lifting mechanism 3. The vertical working range of the robotic arm can be expanded through the scissor lifting mechanism 3.

[0054] like Figure 3 As shown, in a preferred embodiment of the present invention, the predictive balance control algorithm steps of the wheel-foot hybrid harvesting robot are as follows:

[0055] First, the terrain data processing module in control unit 2 fuses the data from lidar 7 and IMU;

[0056] The processed data is divided into two paths: one path performs ground unevenness analysis to assess the flatness of the area to be stepped on; the other path performs balance state assessment to determine the current tilt and stability of the vehicle body.

[0057] Both sets of information are fed into a four-leg adjustment algorithm, which is based on a predictive control model and calculates the amount of joint adjustment required for each of the four legs to maintain or restore balance.

[0058] Finally, the joint control calculation module in control unit 2 converts these adjustments into specific control commands for the servo motors of each leg, driving the robot to complete the posture adjustment. The entire process forms a closed loop, and the actual posture of the wheel-leg dual-mode mobile chassis 1 is fed back to control unit 2 via IMU, achieving continuous and stable balance control.

[0059] like Figure 4 As shown, in a preferred embodiment of the present invention, the 2.5D terrain mapping steps of the wheel-foot hybrid harvesting robot are as follows:

[0060] To achieve intelligent path planning and mode switching, the robot needs an accurate understanding of the surrounding terrain.

[0061] First, the raw point cloud data collected by LiDAR 7 is preprocessed (filtering and noise reduction) and coordinate unified (sensor → world coordinate system). Then, the ground is segmented using an algorithm to extract the pure terrain surface point cloud.

[0062] Subsequently, the control unit 2 performs gridding (dividing the terrain into grids and calculating the average elevation of each grid) and elevation interpolation (filling in the data in sparse areas of the point cloud) in parallel, and then merges the two to generate a DEM.

[0063] Finally, by analyzing the DEM, key terrain features such as slope and ruggedness are extracted to form a 2.5D terrain map for use in subsequent motion planning and pattern decision-making modules. Ruggedness can be quantified using indicators such as TRI.

[0064] like Figure 5 As shown, in a preferred embodiment of the present invention, the multi-mode motion control steps of the wheel-leg hybrid harvesting robot are as follows:

[0065] The motion mode decision module in the control unit receives the calculated TRI value, slope information, and real-time obstacle detection results from the 2.5D terrain map.

[0066] When the robot detects that the terrain ahead is flat (e.g., TRI < 0.01) and unobstructed, it switches to high-speed wheeled mode. In this mode, the hub drive motor 9 drives the wheels 10 to move the robot quickly, while the leg joints maintain a fixed posture, resulting in low energy consumption and high efficiency.

[0067] When a slightly uneven terrain is detected (e.g., 0.01 ≤ TRI < 0.05), the robot switches to a wheeled + active suspension mode. In this mode, the robot is still driven by wheels 10, but the leg joints extend and retract at a small range and high frequency according to a predictive balance control algorithm to absorb ground impacts and act as an active suspension to ensure platform stability.

[0068] When the robot senses rugged terrain or needs to cross obstacles (e.g., TRI ≥ 0.05), it switches to a slow-moving legged mode. In this mode, the hub drive motor 9 locks to secure the wheels 10, and the robot slowly traverses the complex area with a stable gait by alternately lifting and moving its four legs.

[0069] When the robot navigates to the designated picking point, it enters leveling mode, extends its four legs for support, and precisely adjusts the leg length to adjust the upper platform to a completely level position, providing a zero-tilt reference for the robotic arm's operation.

[0070] In any mode, if the sensors detect an emergency obstacle, the emergency obstacle avoidance mode will be triggered, and the robot will immediately stop or perform a detour.

[0071] like Figure 6 As shown, in a preferred embodiment of the present invention, the collaborative precision positioning steps of the wheel-foot hybrid harvesting robot are as follows:

[0072] To achieve precise harvesting, the robot must know its exact location and posture within the orchard. Therefore, a multi-sensor fusion localization scheme is employed. The SLAM module of the LiDAR 7 is responsible for building a global map over a large area and providing a coarse, low-drift pose estimate. Simultaneously, the visual odometry (VO) module of the depth camera 5 provides high-frequency, high-precision relative motion estimation at close range. Attitude data from the IMU, LiDAR pose, and VO pose are input together into the Extended Kalman Filter (EKF) fusion module in the control unit 2. The EKF fusion module combines the advantages of each sensor, suppressing noise and drift, and ultimately outputs a high-precision, high-frequency (>100Hz) 6-DOF pose for the six-axis robotic arm 4 to perform coordinate transformations and precisely locate the harvesting target.

[0073] As a preferred embodiment of the present invention, the specific calculation process of the predictive balance control algorithm is as follows:

[0074] 1) Ground unevenness analysis and calculation:

[0075] For the terrain point cloud data ahead, the least squares method is used for surface fitting;

[0076] The equation of the surface is: ;

[0077] By solving the least squares problem ;

[0078] The final surface coefficient matrix is ​​obtained as follows: ;

[0079] in: Points in the terrain point cloud in the world coordinate system The three-dimensional coordinates in and The horizontal coordinate is... These are elevation coordinates. For the first The coordinates of a terrain point in the world coordinate system. The coefficients are the surface fitting coefficients, which are obtained by solving the least squares method. This is the surface coefficient matrix, containing all fitting coefficients.

[0080] 2) Curvature calculation:

[0081] Calculate the average curvature based on the fitted surface:

[0082] ;

[0083] The first and second partial derivatives involved in the formula are derived from the terrain fitting surface equation. The calculation yields the following specific definition:

[0084] ;

[0085] ;

[0086] ;

[0087] ;

[0088] ;

[0089] in, The mean curvature is used to describe the macroscopic curvature of the terrain surface. and For the terrain surface in and The first partial derivative of the direction, physically representing the slope of that point on the terrain surface. , and It is the second-order partial derivative of the terrain surface, describing the rate of slope change, and is the core component for calculating curvature.

[0090] 3) Zero Moment Point (ZMP) Stability Calculation:

[0091] ;

[0092] ;

[0093] in: For the first The ground reaction vector of a leg. , and For the first The ground reaction force of one leg , and The directional component. For the first The foot of one leg in the robot coordinate system The coordinates in the diagram. The coordinates of the zero torque point in the robot coordinate system.

[0094] 4) Calculation of adjustment amount for four legs:

[0095] Solve using the quadratic programming method:

[0096] ;

[0097] in: For the joint angle adjustment vector; This is the ground reaction force adjustment vector; Adjust the weight matrix for the joints. The weighting matrix for ground reaction force was adjusted and calibrated experimentally. To optimize the objective function.

[0098] The constraints are as follows:

[0099] Dynamic constraints: This constraint is an approximation of the robot's dynamics equations based on the linearized current state.

[0100] Stability constraints: ;

[0101] The supporting polygon refers to the smallest convex hull region formed by connecting all the foot positions in the ground state, ensuring... Being located within this area is a necessary condition for the robot's dynamic stability; This is the quality matrix; Here is the damping matrix; Here is the stiffness matrix;

[0102] Joint limitation: ;

[0103] in, This is the current joint angle vector; and These are the minimum and maximum limits of the joint angle.

[0104] As a preferred embodiment of the present invention, the specific calculation process for 2.5D terrain mapping is as follows:

[0105] 1) Ground segmentation algorithm:

[0106] RANSAC plane fitting was used:

[0107] For a point cloud P, randomly select 3 points to calculate the plane equation:

[0108] ;

[0109] Calculate the distance from all points to the plane. :

[0110] ;

[0111] Interior point set: ;

[0112] in: These are the coefficients of the plane equation. For the first The distance from a point to the plane. The distance threshold (m) for determining interior points in RANSAC is usually determined experimentally or empirically.

[0113] Repeat the iteration, selecting the plane with the most interior points as the ground.

[0114] 2) DEM generation:

[0115] Grid division: The terrain is divided into a grid of 0.1m × 0.1m;

[0116] Elevation interpolation: Inverse distance weighting (IDW) method is used.

[0117] ;

[0118] in: This is the weighting coefficient, which is inversely proportional to the square of the distance. For the first Data points to the grid center The distance. For grid center The elevation interpolation results at the location are based on point cloud data in the world coordinate system. Calculated.

[0119] 3) TRI Calculation: For each grid cell, calculate its elevation change relative to its 8 neighboring grids to assess terrain ruggedness.

[0120] ;

[0121] The specific calculation process is as follows:

[0122] ;

[0123] ;

[0124] but:

[0125] ;

[0126] in: The index coordinates of the current grid cell; For the current grid cell Elevation value; for to , is the elevation value of the 8 neighboring grid cells of the current grid cell; to This represents the absolute value of the elevation difference between the current grid cell and its neighboring grid cells; For the current grid cell The terrain ruggedness index.

[0127] 4) Slope calculation:

[0128] Calculate slope based on DEM data:

[0129] ;

[0130] in: Elevation change rate in direction ; Elevation change rate in direction ; and For grid in and The dimension of the direction, here is ; This represents the slope value.

[0131] In a preferred embodiment of the present invention, the multi-mode motion control threshold is calculated as follows:

[0132] 1) Basis for determining the TRI threshold:

[0133] Calculation of robot's physical parameters:

[0134] Wheeled mode can be driven by:

[0135] ;

[0136] Active suspension compensation range:

[0137] ;

[0138] Foot-based stance height:

[0139] ;

[0140] in: The height that a wheeled vehicle can pass through is determined by the wheel radius. Decide; This represents the maximum compensation range of the active suspension system. This represents the maximum span height in the foot-based mode. Thigh length ; Calf length ; and For the angles of the hip and knee joints, take [the value here]. .

[0141] 2) Mode switching logic:

[0142] If TRI < 0.01 and there is no accessibility:

[0143] Wheeled high-speed mode: =1.0m / s

[0144] elif 0.01≤TRI<0.05:

[0145] Wheel type + active suspension: v=0.5m / s, active suspension frequency f=2-5Hz

[0146] elif TRI≥0.05:

[0147] Slow-paced leg movement: v = 0.2 m / s, gait cycle T = 2 s

[0148] in: This is the maximum speed in wheel mode; This represents the movement speed in the current mode. For active suspension frequency; The gait cycle is in the foot-based mode.

[0149] The TRI threshold mentioned above is a typical value calculated based on the specific parameters of the robot in this embodiment. In practical applications, it can be precisely calibrated through experiments according to the physical size, weight distribution, and specific working environment of different robots.

[0150] In a preferred embodiment of the present invention, the calculation process for cooperative precise positioning is as follows:

[0151] 1) EKF state model:

[0152] State vector: ;

[0153] Process model: ;

[0154] Observation model: ;

[0155] in: The coordinates of the robot's position in the world coordinate system are in meters (m). This represents the robot's velocity components in the world coordinate system, expressed in meters per second (m / s). Let be the robot's pose quaternion; To control the input vector; The noise is a process noise that follows a Gaussian distribution. ; To observe the noise, it follows a Gaussian distribution. ; The process noise covariance matrix is... To observe the noise covariance matrix.

[0156] This is the system state transition function (which may be nonlinear); The observation function maps the state to the sensor measurement space.

[0157] 2) EKF prediction steps:

[0158] ;

[0159] ;

[0160] in: For a moment Prior state estimation, To estimate the covariance matrix a priori, Let be the Jacobian matrix of the state transition function.

[0161] 3) EKF update steps:

[0162] ;

[0163] ;

[0164] ;

[0165] in: The Kalman gain matrix; Let be the Jacobian matrix of the observation function; It is an identity matrix.

[0166] 4) Sensor data time synchronization:

[0167] By employing interpolation methods combined with hardware timestamps, the differences in timestamps between data acquisition and transmission from different sensors are addressed, ensuring temporal consistency in data fusion.

[0168] For time The required data, if :

[0169] ;

[0170] in: For a moment Interpolated data. and Adjacent timestamps and Sensor data.

[0171] In a preferred embodiment of the present invention, the kinematic calculation is as follows:

[0172] 1) Inverse kinematics calculation:

[0173] Given foot position Calculate the joint angles:

[0174] ;

[0175] ;

[0176] ;

[0177] ;

[0178] ;

[0179] ;

[0180] ;

[0181] in: The coordinates of the foot in the hip joint coordinate system; The hip joint yaw angle; The thigh pitch angle; The knee joint pitch angle; The distance is the projection of the foot onto the horizontal plane. The straight-line distance from the hip joint to the foot; The angle between the foot vector and the horizontal plane; It is the angle between the thigh and the line connecting the hip joint and the foot.

[0182] 2) Jacobian matrix calculation:

[0183] Used for speed mapping and force control:

[0184] ;

[0185] Joint velocity to foot velocity: ;

[0186] Foot force to joint torque: ;

[0187] in: The Jacobian matrix of the mechanical leg is a Matrix; : Foot linear velocity vector; This is the joint angular velocity vector; This is the joint torque vector; This is the force vector acting on the foot.

[0188] In a preferred embodiment of the present invention, the algorithm parameter calibration steps are as follows:

[0189] Based on the robot's actual mechanical parameters: thigh length ; calf length Wheel radius Total mass Single-leg weight Maximum motor torque ;

[0190] Control parameters calibrated through experiments:

[0191] Joint adjustment weight matrix: ;

[0192] Ground reaction force adjustment weight matrix: ;

[0193] EKF process noise covariance :

[0194] ;

[0195] EKF observation noise covariance :

[0196] ;

[0197] The algorithm's performance was verified through simulation and experiments.

[0198] Balance control accuracy: On a 10° slope, the platform tilt angle control error is < 0.5°;

[0199] Positioning accuracy: Absolute positioning error < 0.05m, relative positioning error < 0.01m;

[0200] Mode switching response: From detecting terrain change to completing mode switching < 0.5s;

[0201] Harvesting success rate: In leveling mode, the robotic arm's harvesting success rate is > 95%.

[0202] In summary, this invention, through its innovative wheel-foot composite structure and hierarchical collaborative intelligent control strategy, successfully automates the entire process of harvesting robots in unstructured agricultural environments, from efficient movement to stable operation, effectively improving operational efficiency and stability in unstructured environments.

[0203] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A wheel-foot hybrid picking robot, characterized in that, The wheel-foot dual-mode mobile chassis, a perception system, and a working module are included. The wheel-foot dual-mode mobile chassis adopts a quadruped bionic layout, each foot end is wrapped by a foot end shell, the bottom of the foot end shell is provided with a wheel, and the inside of the foot end shell is provided with a hub drive motor for driving the wheel to rotate, the hub drive motor has a servo locking function, when it is needed to switch to the foot mode, the controller of the hub drive motor will make the hub drive motor enter a large damping or locking state, so as to fix the wheel and make the foot end become a stable support point. The perception system includes a laser radar, a depth camera, and a control unit, the laser radar is installed at the front end of the wheel-foot dual-mode mobile chassis, is used for long-distance environment scanning and SLAM mapping, the depth camera is installed at the execution end of the working module, is used for close-range fruit identification, positioning, and fine navigation, and the control unit is arranged inside the wheel-foot dual-mode mobile chassis, is used for processing sensor data and executing decision and control algorithms, and the control unit is internally provided with an IMU, is used for acquiring real-time attitude information of the robot. The working module is installed on the wheel-foot dual-mode mobile chassis, is used for picking work. The multi-mode motion control steps of the wheel-foot composite picking robot are as follows: The motion mode decision module in the control unit receives the calculated TRI value, slope information and real-time obstacle detection result from the 2.5D terrain map; When the robot perceives that the front terrain is flat, i.e. TRI<0.01, and there is no obstacle, switch to the wheeled high-speed mode; in this mode, the hub drive motor drives the wheel to move the robot, and the leg joints remain in a fixed posture; When a slight uneven terrain is perceived, i.e. 0.01≤TRI<0.05, switch to the wheeled+active suspension mode; in this mode, the robot is still driven by the wheel, but the leg joints will stretch and contract in a small range and at a high frequency according to the predicted balance control algorithm to absorb the ground impact and play the role of active suspension; When rugged terrain or obstacles need to be crossed, i.e. TRI≥0.05, switch to the foot slow-speed mode; in this mode, the hub drive motor enters the locking state to fix the wheel, and the robot passes through the complex area by alternately lifting and moving the four legs in a gait; When navigating to the predetermined picking point, the robot enters the leveling mode to adjust the upper platform to be completely horizontal, providing a zero-inclination reference for the mechanical arm operation; In any mode, if an emergency obstacle is detected by the sensor, the emergency obstacle avoidance mode is triggered, and the robot will stop immediately or perform a detour action.

2. The wheel-foot hybrid picking robot according to claim 1, characterized in that, The working module includes a six-axis mechanical arm, the six-axis mechanical arm is installed on the wheel-foot dual-mode mobile chassis, a gripper end effector for picking fruits is installed at the end of the six-axis mechanical arm, and the depth camera is also installed at the end of the six-axis mechanical arm.

3. The wheel-foot hybrid picking robot according to claim 2, wherein, A scissor lifting mechanism is arranged on the wheel-foot dual-mode mobile chassis, and the six-axis mechanical arm is installed on the wheel-foot dual-mode mobile chassis through the scissor lifting mechanism.

4. The wheel-foot hybrid picking robot according to any one of claims 1-3, characterized in that, The steps of the predicted balance control algorithm of the wheel-foot composite picking robot are as follows: First, the terrain data processing module in the control unit fuses the data of the laser radar and the IMU; The processed data is divided into two paths: one path is for ground unevenness analysis to evaluate the flatness of the area to be stepped on; the other path is for balance state evaluation to judge the inclination and stability of the current vehicle body; Both pieces of information are input into the four-legged adjustment algorithm, which is based on a predictive control model to calculate the joint adjustment amount of each of the four legs required to maintain or restore balance; Finally, the adjustment amount is converted into specific control instructions for each leg servo motor through the joint control calculation module in the control unit to drive the robot to complete the posture adjustment; The whole process forms a closed loop, and the actual posture of the wheel-foot dual-mode mobile chassis is fed back to the control unit through the IMU for continuous balance control.

5. The wheel-foot hybrid picking robot according to claim 4, characterized in that, The 2.5D terrain mapping steps of the wheel-foot composite picking robot are as follows: First, after preprocessing and coordinate unification of the original point cloud data collected by the laser radar, the ground is segmented through an algorithm to extract pure terrain surface point cloud; Subsequently, the control unit performs grid processing and elevation interpolation in parallel, and fuses the two to generate a DEM; Finally, by analyzing the DEM, key terrain features including slope and ruggedness are extracted to form a 2.5D terrain map; the ruggedness is quantified by the TRI index, which can be adjusted according to the specific application scenario and physical parameters of the robot.

6. The wheel-foot hybrid picking robot according to claim 5, wherein, The cooperative fine positioning steps of the wheel-foot composite picking robot are as follows: The SLAM module of the laser radar is responsible for constructing a global map and providing pose estimation in a large range; at the same time, the visual odometry module of the depth camera provides relative motion estimation at close range; The attitude data from the IMU, the laser radar pose, and the visual odometry pose are input into the EKF fusion module in the control unit; The EKF fusion module combines the data from each sensor to suppress noise and drift, and finally outputs a 6-DOF pose for the 6-axis robot arm to perform coordinate transformation and fine positioning of the picking target.

Images