Intelligent self-adaptive exoskeleton robot multi-view sensing obstacle avoidance method and system

Abstract

The invention relates to the field of exoskeleton robot technology and robot control, in particular to an intelligent self-adaptive exoskeleton robot multi-view sensing obstacle avoidance method and system. A binocular camera is combined with a laser radar to monitor the external environment, and when it is detected that an obstacle exists around a wearer, the obstacle avoidance is performed; and path searching is carried out by using an improved fast expansion random tree star algorithm, a control instruction is generated based on exoskeleton kinematics modeling, and the exoskeleton robot is driven to realize an obstacle avoidance function. The obstacle avoidance method can flexibly deal with various complex conditions in a dynamically changing environment, is high in real-time performance, can help a wearer to complete obstacle avoidance in some emergency conditions, fully reflects the intelligent self-adaptive capability of the system, and enhances the safety and reliability.

Description

Technical Field

[0001] The present invention relates to the field of exoskeleton robot technology and robot control, and particularly to an intelligent adaptive multi-camera perception obstacle avoidance method and system for an exoskeleton robot. Background Art

[0002] As a wearable intelligent device, exoskeleton robots have been widely used in the fields of medical rehabilitation, industrial assistance, and military in recent years. In the currently widely used technologies, the obstacle avoidance of exoskeleton robots mainly relies on a single sensor. However, lidar is prone to failure in detecting blind spots at close range or transparent obstacles. Although binocular vision is superior to monocular vision, depth estimation is prone to deviation due to the influence of light changes and texture loss, making it difficult to adapt well to dynamic and complex environments. Existing path planning algorithms are difficult to balance computational efficiency and path optimality. Static parameter settings are difficult to cope with sudden obstacles or narrow channels, and the lack of semantic information fusion leads to a single obstacle avoidance strategy. Traditional systems are independent of the user's motion intention, and it is easy to cause the wearer to lose balance due to obstacle avoidance actions. The dynamic load change is not incorporated into the control model, seriously affecting safety and the user experience.

[0003] Especially for the elderly group with limited mobility, the existing technologies have the following deficiencies: 1. The elderly have poor reaction abilities, and the traditional obstacle avoidance system of exoskeleton robots cannot trigger an emergency stop in a timely manner in the scenario of suddenly appearing obstacles, resulting in a high risk of collision; 2. Existing exoskeleton robots lack the ability to sense backward, and cannot detect obstacles approaching from behind. When the elderly are backing up or turning around, due to the decline in body coordination and reaction speed, they are more likely to lose balance and fall due to not noticing the obstacles behind. Therefore, to solve these problems, it is very necessary to study an intelligent adaptive exoskeleton robot that can combine multi-camera perception obstacle avoidance functions. Summary of the Invention

[0004] Aiming at the problems of strong limitations of a single sensor, difficulty for path planning algorithms to quickly and accurately plan the best path, and insufficient human-machine interaction ability in the prior art, the present invention provides an intelligent adaptive multi-camera perception obstacle avoidance method and system for an exoskeleton robot. This method helps the wearer efficiently complete the obstacle avoidance task through multi-sensor fusion, an improved path planning algorithm, and the optimization of human-machine cooperation.

[0005] To achieve the intention of the present invention, the present invention presents the following technical solutions:

[0006] An intelligent adaptive multi-camera perception obstacle avoidance method for an exoskeleton robot, comprising the following steps:

[0007] Step 1: A binocular camera and a lidar obtain multi-modal perception data around the wearer in real time;

[0008] Step 2: Jointly calibrate the extrinsic parameters of the binocular camera and the lidar, and transform the data to the same coordinate system through coordinate transformation;

[0009] Step 3: Register and fuse the visual point cloud of the binocular camera and the 3D point cloud of the lidar to generate a dense environment map;

[0010] Step 4: Based on the improved RRT* algorithm, dynamically adjust the path planning parameters in combination with the environmental map complexity to generate an obstacle avoidance path;

[0011] Step 5: Through the kinematic modeling of the exoskeleton robot, convert the obstacle avoidance path into joint control instructions;

[0012] Step 6: Use the CAN bus protocol to send the control instructions to the drive unit, and the exoskeleton robot realizes the obstacle avoidance function.

[0013] Preferably, in step 1, the binocular camera collects the visual image information of the obstacle, and the lidar collects the spatial position information of the obstacle.

[0014] Preferably, in step 2, the jointly calibrated extrinsic parameters are used to transform the coordinate systems of the binocular camera and the lidar to a consistent coordinate system. The coordinate transformation includes rigid body transformation, perspective transformation and affine transformation, and finally transforms to the pixel coordinate system.

[0015] Specifically, in the joint calibration, the plane equations of the calibration board in different coordinate systems are determined as:

[0016]

[0017] where a m,i and a l,i are the plane equations of the calibration board in the coordinate systems of the binocular camera and the lidar. To obtain the [R|T] matrix from the lidar coordinate system to the camera coordinate system, a minimum loss function can be constructed as follows:

[0018]

[0019] Preferably, in step 3, the visual point cloud of the binocular camera and the 3D point cloud of the lidar are registered and fused to generate a dense environment map, which specifically includes the following steps:

[0020] Step 3.1: Feature extraction: Use the ORB algorithm to extract the local features of the binocular camera point cloud, and extract the feature points of the lidar point cloud based on curvature.

[0021] Step 3.2: Feature matching: Based on the distance metric method, set a distance threshold. When the distance between two feature descriptors is less than this threshold, feature matching is performed.

[0022] Step 3.3: Transformation Optimization: Estimate the transformation parameters from the lidar point cloud to the binocular camera point cloud through the Iterative Closest Point (ICP) algorithm, with the registration error ≤ 1 cm. After multiple iterations, accurate transformation parameters are obtained to achieve precise alignment of the two point clouds.

[0023] Step 3.4: Data Fusion: Adopt a weighted average strategy for data fusion. For areas with short distances or obvious texture features, increase the weight of the binocular camera point cloud; for areas with long distances or high requirements for spatial position accuracy, increase the weight of the lidar point cloud.

[0024] Preferably, in Step 3.4, the weighted average strategy includes a weight calculation formula, specifically:

[0025]

[0026] ω2 = 1 - ω1;

[0027] In the formula, ω1 is the weight of the binocular camera, ω2 is the weight of the lidar, d0 = 3 m is the set threshold, and k is the adjustment coefficient. When the distance d of the obstacle < 3 m, the weight of the binocular camera ω1 ≥ 70%, and when the distance d of the obstacle ≥ 3 m, the weight of the lidar ω2 ≥ 60%.

[0028] Preferably, in Step 4, for the improved RRT* algorithm, a goal biasing strategy is introduced to include the goal point in the selection range of the sampling points. By setting the biasing probability P, the convergence of path planning is significantly accelerated. The information entropy is used to measure the complexity of the map, and P and the search step size S are dynamically adjusted. The map complexity can be calculated by the information entropy formula:

[0029]

[0030] In the formula, H represents the information entropy, which includes the distance factor and the speed of dynamic obstacles. The larger H is, the higher the map complexity. λ is the dynamic weight factor, λ ∈ [0.1, 0.3], and it can be adaptively adjusted according to the scene complexity. When the map complexity is high, the values of the biasing probability P and the step size S should be reduced.

[0031]

[0032] In the formula, the biasing probability P ∈ [0.1, 0.3], and the step size S does not exceed 0.1 - 0.3 times the length of the map diagonal.

[0033] Preferably, in Step 5, kinematic modeling is performed on the exoskeleton robot, specifically including the following steps:

[0034] Step 5.1: The optimal obstacle avoidance path is discretized into a key point sequence at dynamically adjusted time intervals. Each joint point includes the end pose X = [x, y, z, θ x,θ y ,θ z ] T . A quintic polynomial interpolation algorithm is used to generate smooth motion trajectories and avoid sudden changes in joint motion.

[0035] Step 5.2: Use analytical or numerical methods to determine the required angles for each joint based on the desired position and orientation of the end effector.

[0036] Step 5.3: Introduce joint angle, angular velocity, and torque constraints to ensure that the solved joint angle is within the physically feasible range of the mechanical structure. Use the IMU to detect the torso posture and dynamically adjust the joint angle and angular velocity to maintain balance.

[0037] Preferably, in step 5.2, for a serial exoskeleton robot with a relatively simple structure and DOF ≤ 3, an analytical method is used; for an exoskeleton robot with a complex structure, DOF ≥ 4 or kinematic redundancy, a numerical method is used.

[0038] Preferably, in step 6, the CAN bus protocol adopts a priority arbitration mechanism to define the joint instruction priority, and sends the coded instructions to the drive unit through the CAN bus in a time-sharing manner according to the joint movement priority. After decoding by the drive unit, the PID control algorithm is used to adjust the actuator to accurately control the joint movement and realize the obstacle avoidance function.

[0039] An intelligent and adaptive multi-eye perception and obstacle avoidance system for an exoskeleton robot includes a binocular camera, a laser radar, a battery pack, an IMU module, and a main control unit. The binocular camera is embedded in a reserved slot in the exoskeleton robot helmet, with the installation position aligned with the human eye's field of view; the laser radar is fixed to the top of the helmet, with a horizontal scanning range covering 360°; the IMU module is arranged at the center of the torso; the main control unit is installed in the back area of the exoskeleton robot; and the battery packs are symmetrically distributed on both sides of the exoskeleton robot's waist.

[0040] Preferably, the helmet is provided with an adjustable fixing device for adapting to different head shapes and reducing movement jitter, and can protect the wearer, and the main control unit is integrated with a heat dissipation structure.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] By integrating a lidar and a binocular camera on the exoskeleton robot, the present invention fully exploits the complementary advantages of multi-modal perception. The lidar provides stable spatial information in long-distance and complex lighting environments, compensating for the lack of depth estimation of the binocular camera in long-distance and low-texture scenes. The binocular camera, relying on dense disparity calculation and image information, achieves millimeter-level resolution at medium distances. Through spatio-temporal calibration and data fusion, the two build a multi-scale environmental model, significantly enhancing the environmental perception ability. The multi-view perception obstacle avoidance system of the exoskeleton robot can quickly identify sudden obstacles and trigger an emergency stop mechanism. Moreover, for the elderly with slow reaction, it has a rear monitoring function, which can detect rear obstacles in real time, filling the rear blind spot of traditional exoskeleton robots, expanding the application range of exoskeletons, and improving the safety of special user groups.

[0043] The present invention applies the improved RRT* algorithm to the exoskeleton robot, introducing a target biasing strategy to dynamically adjust the biasing probability P and the search step size S. Compared with traditional path planning algorithms that are difficult to balance computational efficiency and path optimality, it improves the planning efficiency and path quality, ensuring the safe and efficient movement of the wearer. Kinematic modeling of the exoskeleton robot is also carried out. During the process of generating obstacle avoidance instructions, human-machine interaction is fully considered, and constraint conditions are introduced to ensure that the solved joint angles are within the physically feasible range of the mechanical structure, improving the user experience of the wearer. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] Figure 1 It is a flowchart of a multi-view perception obstacle avoidance method for an intelligent adaptive exoskeleton robot of the present invention. Figure 2 It is a schematic diagram of the installation position of the components of the multi-view perception obstacle avoidance system.

[0045] Figure 3 It is a flowchart of the improved RRT* algorithm of the present invention.

[0046] Figure 4 It is a schematic diagram of the multi-view perception obstacle avoidance process of an intelligent adaptive exoskeleton robot. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0047] The following further elaborates on the present invention by combining the embodiments and the attached Figures 1 to 3 It should be clear that the embodiments described here are only part of the embodiments of the present invention, not all of them. Based on the embodiments provided by the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the scope of protection of the present invention.

[0048] Embodiment 1, as Figure 1 , a multi-view perception obstacle avoidance method for an intelligent adaptive exoskeleton robot, comprising the following steps:

[0049] Step 1: The binocular camera and lidar acquire multi-modal perception data around the wearer in real time;

[0050] Step 2: Jointly calibrate the extrinsic parameters of the binocular camera and lidar, and transform the data to the same coordinate system through coordinate transformation;

[0051] Step 3: Register and fuse the visual point cloud of the binocular camera and the 3D point cloud of the lidar to generate a dense environment map;

[0052] Step 4: Based on the improved RRT* algorithm, dynamically adjust the path planning parameters in combination with the environmental map complexity to generate an obstacle avoidance path;

[0053] Step 5: Through the kinematic modeling of the exoskeleton robot, convert the obstacle avoidance path into joint control instructions;

[0054] Step 6: Use the CAN bus protocol to send the control instructions to the drive unit, and the exoskeleton robot realizes the obstacle avoidance function.

[0055] In this embodiment, as Figure 2 shown, the multi-view perception system of the exoskeleton robot includes a lidar 1, a binocular camera 2, an IMU module 3, a battery pack 4, and a main control unit 5 integrated. The lidar is fixed on the top of the helmet, and the horizontal scanning range covers 360°, which can cover the surrounding environment of the whole body; the binocular camera 2 is embedded in the reserved slot of the exoskeleton robot helmet, and the installation position is flush with the human eye view angle, which can enhance the intuitiveness of obstacle avoidance decision-making; the IMU module 3 is arranged at the center position of the trunk, close to the center of gravity of the human body, which can accurately reflect the overall motion state; the main control unit 5 is installed in the back area of the exoskeleton robot, where there is sufficient space; the battery packs 4 are symmetrically distributed on both sides of the waist of the exoskeleton robot to reduce wearing fatigue.

[0056] In this embodiment, the helmet is provided with an adjustable fixing device for adapting to different head shapes and reducing motion jitter, and can protect the wearer, and the main control unit is integrated with a heat dissipation structure.

[0057] In this embodiment, in step 1, the binocular camera collects visual image information of the obstacle, and the lidar collects spatial position information of the obstacle.

[0058] As Figure 3 shown, the exoskeleton robot collects information about the obstacle in front through the binocular camera and lidar, and detects a suddenly emerging moving obstacle in front.

[0059] In this embodiment, in step ⒉, the joint calibration of the extrinsic parameters is used to transform the coordinate systems of different sensors such as the binocular camera and lidar to a consistent coordinate system. The coordinate transformation includes rigid body transformation, perspective transformation and affine transformation, and finally transforms to the pixel coordinate system. Specifically, the plane equations of the calibration board in different coordinate systems are determined as:

[0060]

[0061] where a m,i and a l,i are the plane equations of the calibration board in the binocular camera and lidar coordinate systems. To obtain the [R|T] matrix from the lidar coordinate system to the camera coordinate system, a minimum loss function can be constructed as follows:

[0062]

[0063] where l(i) represents the number of points on the calibration board plane in the i-th frame, and j represents the serial number of the points in each frame.

[0064] In this embodiment, in step 3, the visual point cloud of the binocular camera and the three-dimensional point cloud of the lidar are registered and fused to generate a dense environment map, which specifically includes the following steps:

[0065] Step 3.1: Feature extraction: Use the ORB algorithm to extract the local features of the binocular camera point cloud, and extract the feature points of the lidar point cloud based on curvature.

[0066] Step 3.2: Feature matching: Based on the distance metric method, set a distance threshold. When the distance between two feature descriptors is less than this threshold, feature matching is performed.

[0067] Step 3.3: Transformation optimization: Estimate the transformation parameters from the lidar point cloud to the binocular camera point cloud through the iterative closest point algorithm, with the registration error ≤ 1 cm. After multiple iterations, accurate transformation parameters are obtained to achieve accurate alignment of the two point clouds.

[0068] Step 3.4: Data fusion: Adopt a weighted average strategy for data fusion. For areas with short distances or obvious texture features, increase the weight of the binocular camera point cloud; for areas with long distances or high requirements for spatial position accuracy, increase the weight of the lidar point cloud.

[0069] In this embodiment, in step 3.4, the weighted average strategy includes a weight calculation formula, specifically:

[0070]

[0071] ω2 = 1 - ω1;

[0072] where ω1 is the weight of the binocular camera, ω2 is the weight of the lidar, d0 = 3 m is the set threshold, and k is the adjustment coefficient. When the distance d of the obstacle < 3 m, the binocular camera weight ω1 ≥ 70%, and when the distance d of the obstacle ≥ 3 m, the lidar weight ω2 ≥ 60%.

[0073] Such as Figure 3, To improve the flowchart of the RRT* algorithm, in this embodiment, in step 4, the improved RRT* algorithm introduces a goal biasing strategy, incorporating the goal point into the selection range of sampling points. By setting the biasing probability P, the convergence of path planning is significantly accelerated. The information entropy is used to measure the complexity of the map, and P and the search step size S are dynamically adjusted. The map complexity can be calculated by the information entropy formula:

[0074]

[0075] In the formula, H represents the information entropy, which includes the distance factor and the speed of dynamic obstacles. The larger H is, the higher the map complexity. λ is the dynamic weight factor, λ ∈ [0.1, 0.3], and it can be adaptively adjusted according to the scene complexity.

[0076] When the map complexity is high, the values of the biasing probability P and the step size S should be decreased.

[0077]

[0078] In the formula, the biasing probability P ∈ [0.1, 0.3], and when the environmental complexity is high, P is directly set to 0.3, and when the emergency stop is triggered, it is directly set to 0.1. The step size S does not exceed 0.1 - 0.3 times the length of the map diagonal.

[0079] In this embodiment, in step 5, through the kinematic modeling of the exoskeleton robot, the obstacle avoidance path is converted into joint control commands, which specifically includes the following steps:

[0080] Step 5.1: The optimal obstacle avoidance path is discretized into a key point sequence at dynamically adjusted time intervals. Each joint point includes the end pose X = [x, y, z, θ x , θ y , θ z T . The quintic polynomial interpolation algorithm is used to generate a smooth motion trajectory to avoid sudden changes in joint motion.

[0081] Step 5.2: The analytical method or the numerical method is selected to solve the angles required for each joint based on the desired position and pose of the end effector.

[0082] Step 5.3: Joint angle, angular velocity, and torque constraints are introduced to ensure that the solved joint angles are within the physically feasible range of the mechanical structure. The trunk pose is detected by the IMU, and the joint angles and angular velocities are dynamically adjusted to maintain balance.

[0083] In this embodiment, in step 5.2, for a serial exoskeleton robot with a relatively simple structure and degrees of freedom ≤ 3, the analytical method is used, and for an exoskeleton robot with a complex structure, degrees of freedom ≥ 4, or kinematic redundancy, the numerical method is used.

[0084] In this embodiment, in step 6, the CAN bus protocol adopts a priority arbitration mechanism to define the priority of joint instructions, and the encoded instructions are sent to the drive unit time-divisionally according to the joint movement priority through the CAN bus. After being decoded by the drive unit, the PID control algorithm is used to adjust the actuator to accurately control the joint movement and achieve the obstacle avoidance function.

[0085] For example Figure 4 , when the wearer wears the exoskeleton robot and walks forward, there is a wall on the left and an obstacle in front. The exoskeleton robot uses the binocular camera and lidar above the head to perceive the external environment. At the same time, the data sensed by the binocular camera and lidar are transmitted to the development board in real time. After the development board fuses and processes the data, the improved RRT* algorithm is used to plan the optimal path. Based on the planned optimal path, the development board combines the attitude information of the exoskeleton robot fed back by the IMU module at the center of the torso, and adjusts the joint motors of the exoskeleton robot through PID control to make it move smoothly to the right side of the obstacle and complete obstacle avoidance.

[0086] Embodiment 2, a multi-camera perception obstacle avoidance system for an intelligent adaptive exoskeleton robot, the multi-camera perception obstacle avoidance system includes a binocular camera, a lidar, a battery pack, an IMU module, and a main control unit. The binocular camera is embedded in the reserved slot of the exoskeleton robot helmet, and the installation position is flush with the human eye view; the lidar is fixed on the top of the helmet, and the horizontal scanning range covers 360°; the IMU module is arranged at the center of the torso; the main control unit is installed in the back area of the exoskeleton robot; the battery packs are symmetrically distributed on both sides of the waist of the exoskeleton robot.

[0087] Preferably, the helmet is provided with an adjustable fixing device for adapting to different head shapes and reducing movement jitter, and can protect the wearer. The main control unit integrates a heat dissipation structure.

[0088] The above only shows one of the numerous preferred embodiments of the present invention, and it is not intended to impose any restrictions on the technical solutions of the present invention. Those skilled in the art should understand that, on the premise of always adhering to the core spirit and basic principles of the present invention, a series of reasonable and simple modifications and replacements can be made to the technical solutions of the present invention. And these modifications and replacements made based on the original solutions are all included in the protection scope defined by the claims.

Claims

1. An intelligent adaptive multi - vision perception obstacle avoidance method for exoskeleton robots, characterized in that It includes the following steps: Step 1: The binocular camera and lidar acquire multi-modal perception data around the wearer in real time; Step 2: Jointly calibrate the extrinsic parameters of the binocular camera and lidar, and transform the data to the same coordinate system through coordinate transformation; Step 3: Register and fuse the visual point cloud of the binocular camera and the 3D point cloud of the lidar to generate a dense environment map; Step 4: Based on the improved Rapidly-Exploring Random Tree Star (RRT*) algorithm, dynamically adjust the path planning parameters in combination with the complexity of the environment map to generate an obstacle avoidance path; Step 5: Through the kinematic modeling of the exoskeleton robot, convert the obstacle avoidance path into joint control instructions; Step 6: Use the CAN bus protocol to send the control instructions to the drive unit, and the exoskeleton robot realizes the obstacle avoidance function.

2. The multi - binocular perception obstacle - avoidance method of an intelligent adaptive exoskeleton robot according to claim 1, characterized in that, In Step 1, the binocular camera collects visual image information of obstacles, and the lidar collects spatial position information of obstacles.

3. An intelligent adaptive multi - vision perception obstacle avoidance method for an exoskeleton robot according to claim 1, characterized in that, In Step 2, the jointly calibrated extrinsic parameters are used to transform the coordinate systems of the binocular camera and lidar to a consistent coordinate system. The coordinate transformation includes rigid body transformation, perspective transformation, and affine transformation, and finally transforms to the pixel coordinate system.

4. For the multi - vision perception obstacle avoidance method of an intelligent adaptive exoskeleton robot described in claim 1, it is characterized in that, In Step 3, the visual point cloud of the binocular camera and the 3D point cloud of the lidar are registered and fused to generate a dense environment map, which specifically includes the following steps: Step 3.1: Feature extraction: Use the ORB algorithm to extract local features of the binocular camera point cloud, and extract feature points of the lidar point cloud based on curvature; Step 3.2: Feature matching: Based on the distance metric method, set a distance threshold. When the distance between two feature descriptors is less than this threshold, feature matching is performed; Step 3.3: Transformation optimization: Estimate the transformation parameters from the lidar point cloud to the binocular camera point cloud through the Iterative Closest Point algorithm, with the registration error ≤ 1 cm. After multiple iterations, obtain accurate transformation parameters to achieve precise alignment of the two point clouds; Step 3.4: Data fusion: Adopt a weighted average strategy for data fusion, and dynamically allocate weights according to the distance of obstacles. When the distance of the obstacle < 3 m, the weight of the binocular camera ≥ 70%, and when the distance ≥ 3 m, the weight of the lidar ≥ 60%.

5. For a multi - vision perception obstacle - avoidance method of an intelligent adaptive exoskeleton robot according to claim 1, it is characterized in that In Step 4, for the improved RRT* algorithm, introduce the goal biasing strategy to include the goal point in the selection range of sampling points. By setting the biasing probability P, significantly accelerate the convergence of path planning. Use information entropy to measure the complexity of the map, and dynamically adjust P and the search step size S. The map complexity can be calculated by the information entropy formula: In the formula, H represents information entropy, which includes the distance factor and the speed of dynamic obstacles. The larger H is, the higher the map complexity. λ is the dynamic weight factor, λ ∈ [0.1, 0.3], and it is adaptively adjusted according to the scene complexity; when the map complexity is high, the values of the biasing probability P and the step size S should be reduced; In the formula, the biasing probability P ∈ [0.1, 0.3], and the step size S does not exceed 0.1 - 0.3 times the length of the map diagonal.

6. The multi - vision perception obstacle - avoidance method for an intelligent adaptive exoskeleton robot according to claim 1, wherein In Step 5, through the kinematic modeling of the exoskeleton robot, convert the obstacle avoidance path into joint control instructions, which specifically includes the following steps: Step 5.1: The optimal obstacle avoidance path is discretized into a key point sequence at dynamically adjusted time intervals, and each joint point contains the end pose X = [x, y, z, θ x , θ y , θ z T ; The quintic polynomial interpolation algorithm is used to generate a smooth motion trajectory to avoid sudden changes in joint motion;​ Step 5.2: Select the analytical method or numerical method, and solve the angles required for each joint based on the desired position and attitude of the end effector; Step 5.3: Introduce joint angle, angular velocity, and torque constraints to ensure that the solved joint angle is within the physically feasible range of the mechanical structure. Use the IMU to detect the torso posture and dynamically adjust the joint angle and angular velocity to maintain balance.

7. An intelligent adaptive multi - vision perception obstacle - avoidance method for an exoskeleton robot according to claim 1, characterized in that, In step 6, the CAN bus protocol adopts a priority arbitration mechanism to define the joint instruction priority, and sends the encoded instructions to the drive unit through the CAN bus according to the joint movement priority. After decoding by the drive unit, the PID control algorithm is used to adjust the actuator to accurately control the joint movement and realize the obstacle avoidance function.

8. An intelligent adaptive multi-camera perception obstacle avoidance system for an exoskeleton robot, characterized in that The multi-eye perception and obstacle avoidance system includes a binocular camera, a laser radar, a battery pack, an IMU module, and a main control unit. The binocular camera is embedded in a reserved slot in the exoskeleton robot helmet, and the installation position is flush with the human eye's field of view; the laser radar is fixed on the top of the helmet, and the horizontal scanning range covers 360°; the IMU module is arranged at the center of the torso; the main control unit is installed in the back area of the exoskeleton robot; and the battery pack is symmetrically distributed on both sides of the exoskeleton robot's waist.

9. The intelligent adaptive multi-camera perception obstacle avoidance system for an exoskeleton robot according to claim 8, wherein The helmet is provided with an adjustable fixing device for adapting to different head shapes and reducing movement jitter, and can protect the wearer. The main control unit is integrated with a heat dissipation structure.

10. The intelligent adaptive multi-camera perception obstacle avoidance system for an exoskeleton robot according to claim 8, wherein, When the wearer wears the exoskeleton robot and is operating, the multi-eye perception and obstacle avoidance system can quickly identify sudden obstacles and trigger the emergency stop mechanism. At the same time, for the elderly with poor reaction, the system has a rear area monitoring function. When it detects that there is an object approaching in the rear area or there is a risk of collision, the system will issue a vibration warning in advance and assist the elderly to make corresponding evasive actions.

Images