A method, robot, and system for autonomous disembarkation of a robot

Abstract

The application provides a method, a robot and a system for autonomous getting-off of a robot, and relates to the technical field of robot control. The method provided by the application comprises the following steps: in response to a getting-off instruction, determining a drivable area from a current position to a door position; controlling a robot body to move towards the door in the drivable area; collecting a first depth image in the direction of the door, and performing door edge detection on the first depth image to obtain the position of a lower edge of the door; and controlling the robot body to get off according to the position of the detected lower edge of the door. The application provides a perfect solution for autonomous getting-off of the robot, enhances the adaptability and autonomy of the robot in various terrains, and thus realizes higher operation reliability and safety.

Description

Technical Field

[0001] This application relates to the field of robot control technology, and in particular to a method, robot, and system for a robot to autonomously disembark. Background Technology

[0002] A robot is an intelligent machine capable of semi-autonomous or fully autonomous operation. Robots can perform tasks such as operations or movement through programming and automatic control.

[0003] With the development of intelligent driving technology, robots can perform operations such as inspection and driving on vehicles. However, after completing their operations, the robots cannot dismount autonomously and require manual removal by personnel, increasing labor costs. Therefore, this application provides a method, robot, and system for an autonomous robot dismounting from a vehicle. Summary of the Invention

[0004] To address the above problems, this application proposes a method, robot, and system for a robot to autonomously disembark from a vehicle, thereby enabling the robot to disembark autonomously.

[0005] In a first aspect, embodiments of this application provide a method for a robot to autonomously disembark, including:

[0006] In response to the disembarkation command, determine the drivable area from the current location to the door location;

[0007] Within the drivable area, the robot body is controlled to move toward the vehicle door;

[0008] Acquire a first depth image in the direction of the car door, and perform door edge detection on the first depth image to obtain the position of the lower edge of the car door;

[0009] The robot body is controlled to get off the vehicle based on the detected position of the lower edge of the door.

[0010] Secondly, embodiments of this application provide a robot, including: a controller and a robot body connected to the controller;

[0011] The robot body is equipped with a sensing device for collecting images of the direction of the car door;

[0012] The controller is used to respond to a disembarkation command by determining a drivable area from the current position to the door position; within the drivable area, controlling the robot body to move towards the door; acquiring a first depth image in the direction of the door, and performing door edge detection on the image to obtain the position of the lower edge of the door; and controlling the robot body to disembark based on the detected position of the lower edge of the door.

[0013] Thirdly, embodiments of this application provide a robot autonomous disembarkation system, comprising: a robot and a vehicle as described in any embodiment; the robot is located inside the vehicle; and the vehicle is used to issue a disembarkation command to the robot.

[0014] Compared with the prior art, this application has the following technical effects:

[0015] 1. In this application, the robot on the vehicle performs environmental perception and localization, determines the drivable area facing the door, and automatically moves to the door; then it detects the door position and controls the robot to get off the vehicle, providing a complete solution for autonomous robot disembarking.

[0016] 2. This application ensures the safety of robot movement by detecting the drivable area and the lower edge of the door, and enhances the robot's adaptability and autonomy in various terrains, thereby achieving higher operational reliability and safety.

[0017] 3. The solution provided in this application mainly relies on the sensing devices on the robot body to make control decisions, avoiding dependence on external signals and improving the independence and flexibility of the system.

[0018] The above description is merely an overview of the technical solution of this application. In order to better understand the technical means of this application, it can be implemented according to the contents of the specification. In order to make the above description and other objects, features and advantages of this application more obvious and understandable, preferred embodiments are provided and described in detail below. Attached Figure Description

[0019] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the scope of this application. Furthermore, the same reference numerals denote the same parts throughout the drawings. In the drawings:

[0020] Figure 1 A flowchart illustrating a method for a robot to autonomously disembark, provided in an embodiment of this application;

[0021] Figure 2 A schematic diagram of a security map provided in an embodiment of this application;

[0022] Figure 3 A flowchart illustrating another method for a robot to autonomously disembark, provided in an embodiment of this application;

[0023] Figure 4 A 2.5D raster image provided for an embodiment of this application;

[0024] Figure 5 A schematic diagram of the sector-shaped region provided in the embodiments of this application;

[0025] Figure 6 This application provides a 2D occupancy grid map for embodiments of the present application.

[0026] Figure 7 This is a schematic diagram of the robot provided in an embodiment of this application. Detailed Implementation

[0027] Exemplary embodiments of the present disclosure will now be described in more detail with reference to the accompanying drawings. While exemplary embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure may be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the scope of the disclosure to those skilled in the art.

[0028] In the description of this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "joining," "fixing," etc., should be interpreted broadly. For example, they can refer to a connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0029] This application provides a method for a robot to autonomously disembark from a vehicle, applicable to situations where an in-vehicle robot needs to disembark autonomously without human assistance. The method provided in this embodiment is executed by a controller, which can be composed of hardware and / or software. The controller can be integrated into the robot body or electrically connected to the robot body.

[0030] See Figure 1 The method provided in this embodiment includes the following steps:

[0031] S110, in response to the disembarkation command, determine the drivable area from the current position to the door position.

[0032] The vehicle door opens and sends a command to the robot to get out, for example, by broadcasting the voice command "get out" through the vehicle's speaker. The robot then captures and interprets the voice command to obtain the exit instruction.

[0033] When the robot receives the disembarkation command, it may not be facing the door, so the door orientation and position need to be determined first. Optionally, in response to the disembarkation command, a second depth image of the robot's surroundings is acquired; based on the depth data in the second depth image, the door orientation and position are determined.

[0034] In some examples, a depth camera is mounted on the robot. The camera lens can be angled slightly upwards to detect most of the door and human features. Due to the limited field of view of the depth camera, the robot can be controlled to rotate in place and continuously take pictures, obtaining multiple second depth images of the robot's surroundings. The depth data in the second depth images represents the distance to the current position. Because the space outside the door is vast, the rate of change of depth data between adjacent pixels can be calculated, and the pixel with the largest rate of change is determined as the door's location. By combining the position and orientation of the depth camera when taking the second depth images, and the door's location in the second depth images, the real-world door's location and orientation can be obtained.

[0035] After determining the location and direction of the car door, the robot body is controlled to face the car door and the drivable area from the current location to the car door location is determined.

[0036] In some examples, there may be obstacles, such as seats, between the current position and the door position, and it is necessary to determine the drivable area where the robot can move, which may be the floor area inside the vehicle.

[0037] S120. Within the drivable area, control the robot body to move toward the door.

[0038] The drivable area is an interconnected area. Within the drivable area, the robot plans the shortest path from the current position to the door position and controls the robot body to move towards the door along the shortest path.

[0039] S130. Acquire a first depth image in the direction of the car door, and perform door edge detection on the first depth image to obtain the position of the lower edge of the car door.

[0040] As the robot moves towards the car door, a depth camera on the robot takes multiple initial depth images in real time, pointing forward. Door edge detection is then performed on each initial depth image to determine the position of the lower edge of the door (primarily the lower edge), allowing the robot to pause at the lower edge of the door and prevent it from falling out of the car.

[0041] S140. Based on the detected position of the lower edge of the car door, control the robot body to get off the car.

[0042] Once the robot reaches the lower edge of the car door, you can get off the car using either of the following two methods.

[0043] The first method is to jump off the vehicle directly: Detect the height of the lower edge of the vehicle door from the ground, for example, using a depth camera or radar. If this height is less than or equal to the robot's maximum jump height, meaning the robot supports jumping at that height, then control the robot to jump off the vehicle from the lower edge of the door.

[0044] Optionally, in addition to the height of the lower edge of the car door from the ground, the robot also needs to assess whether the disembarkation area meets the following conditions by detecting depth images: whether the ground is flat and whether there are obstacles on the ground.

[0045] The second method involves disembarking via a lower platform, which rests between the lower edge of the door and the ground. A third depth image is captured outside the door using a depth camera. Edge detection of the lower platform is then performed on the third depth image to determine its position. The drivable area is combined with the position of the lower platform edge to obtain a safety map. Within the area corresponding to the safety map, the robot is controlled to disembark along the lower platform.

[0046] The second method hinges on accurately identifying the position of the lower platform edge. Optionally, firstly, the third depth image is filtered to remove noise, for example, using median filtering or Gaussian filtering for smoothing. Then, the depth data in the third depth image is normalized to a uniform range to facilitate subsequent processing, as shown in the following formula:

[0047] ;

[0048] Specifically, D represents the depth data in the third depth image. max It is the maximum depth value in the third depth image. D min It is the minimum depth value in the third depth image. It is normalized depth data.

[0049] Then, the Sobel operator is used to compute the gradient of the third-depth image to detect edges. The Sobel operator can compute the gradient of the image in the x and y directions:

[0050] ;

[0051] Where G represents the gradient magnitude, G x G y This represents the Sobel operator.

[0052] To accurately locate the edges, non-maximum suppression (NMS) is used to preserve local maxima. Finally, the detected undercarriage edges are combined with the drivable area to generate the final safety map. Specifically, the undercarriage edges can be marked on the boundaries of the drivable area to obtain the safety map. See the safety map section. Figure 2As shown, the map is formed by stitching together the edge of the dismount platform (red) and the drivable area (green). Projecting the safety map onto the real world reveals the robot's drivable area. Autonomous dismounting is then achieved within this drivable area.

[0053] In practice, as the robot moves, the third depth image and edge detection results need to be dynamically updated, and detection parameters need to be adjusted to cope with environmental changes. Filters (such as Kalman filters) can be used to smooth the detection results and track edge positions. This embodiment achieves precise disembarkation by identifying the position of the disembarkation platform edge.

[0054] To achieve precise vehicle disembarkation, this embodiment combines drivable area and edge detection technologies. By using a depth camera to detect the lower edge of the door and the edge of the lower panel, and fusing this edge information with drivable area data, a more accurate autonomous disembarkation function can be achieved, improving the safety and accuracy of disembarkation.

[0055] Compared with the prior art, the embodiments of this application also have the following technical effects:

[0056] The onboard robot performs environmental perception and localization, determines the drivable area facing the door, and automatically moves to the door. It then detects the door's position and controls the robot to disembark, providing a complete solution for autonomous robot disembarkation. This embodiment ensures robot safety by detecting the drivable area and the lower edge of the door, enhancing the robot's adaptability and autonomy in various terrains, thereby achieving higher operational reliability and safety. The solution provided in this embodiment primarily relies on the sensing devices on the robot itself for control decisions, avoiding dependence on external signals and improving the system's independence and flexibility.

[0057] This application also provides a method for a robot to autonomously disembark, see [link to relevant documentation]. Figure 3 The detection algorithms for the lower edge of the car door and the algorithms for determining the drivable area are refined to improve detection accuracy. Specifically, the following operations are performed:

[0058] S210: In response to the disembarkation command, collect point cloud data from the current position to the door position.

[0059] The robot uses a lidar on its body to collect point cloud data from its current position to the door position. The point cloud data is composed of the three-dimensional coordinates of each point in the environment captured by the lidar, and has high precision and detailed spatial information. Each point cloud is represented by coordinates (x, y, z) in the lidar coordinate system, with the lidar as the origin, z representing the height of the point cloud in the vertical direction of the lidar, y representing the position of the point cloud in the horizontal direction of the lidar, and x representing the position of the point cloud in the vertical direction of the lidar.

[0060] S220. Perform ground segmentation on the point cloud data to obtain ground point cloud data, which constitutes the drivable area.

[0061] Some point cloud data consists of ground point clouds, which the robot can move through; in this case, the drivable area is constructed using ground point cloud data. Other point cloud data consists of obstacle point clouds, which the robot cannot move through. Therefore, it is necessary to segment the point cloud data into ground point cloud data and obstacle point cloud data. An optional implementation method for ground segmentation is provided below, including steps S221-S224.

[0062] S221. The point cloud data is rasterized to obtain a raster map, wherein each grid in the raster map includes joint height data of multiple point cloud data.

[0063] The point cloud data acquired by the S210 is rasterized into a 2.5D raster map. This raster map divides the environment into regular grids and records the height information in each grid, thus describing the terrain undulations. The purpose of rasterization is to transform complex point cloud data into a regular, easily processed grid format, making the environmental information more structured and simplified. Through rasterization, the point cloud data is divided into uniform grid cells (i.e., graticules), with each grate recording specific environmental features (such as obstacles or ground), making subsequent map processing, analysis, and obstacle avoidance calculations more efficient and intuitive.

[0064] First, choose a suitable grid size (usually using...). d x and d y This means that in this scheme, a value of 0.05 is used, which is equivalent to each grid cell being a 5cm square. The size of the grid cell affects the resolution of the final raster image. Each point cloud in the point cloud data... Mapped to a 2D raster, the index of the point cloud's position within the raster can be calculated using the following formula ( , ), used to normalize point clouds:

[0065] ;

[0066] ;

[0067] in, and It represents the minimum x and y coordinates in the point cloud data, where i is the point cloud number.

[0068] Then, the z-coordinate values ​​of the point clouds are mapped to corresponding 2D graticets. Each grate will contain the z-values ​​of multiple point clouds, and the joint height data (e.g., maximum or average height) of the point clouds within that grate can be calculated to represent the terrain height of that grate. The final 2.5D grate map is shown below. Figure 4 As shown, each color block represents a raster, with different colors indicating different joint height data / terrain height data, denoted as . .

[0069] S222. Map each grid in the grid diagram to multiple sector regions, each sector region including multiple containers, each container mapping the combined height data of multiple grids.

[0070] After obtaining the 2.5D raster image, ground segmentation is required. This is typically achieved by analyzing the height and position of the point cloud, with the aim of removing obstacle information and preserving ground information. To suit applications where the vehicle interior floor has a certain slope, this embodiment provides a ground segmentation method based on ground line fitting. First, see... Figure 5 As shown, the circle is divided into multiple sector regions (i.e., N segments), each segment has an angle of Δα, and each segment is further divided into M equal containers (Bin). Each container is represented as segments_ (bin). Each segment is treated as an independent processing unit for line fitting.

[0071] Project each grid cell into segments_ (bin). That is, project grid P( , , ) Dimensionality reduction to P'(d, ), where d is the position of the grid in the ground coordinate system.

[0072] S223. Determine the minimum integrated height information of each container in each sector region, and perform piecewise linear fitting on the minimum integrated height information of each container in each sector region to obtain the ground line.

[0073] Within each container, the lowest point of elevation is selected. A ground line is fitted based on these lowest points within a sector-shaped area. This line is then used to determine whether a grid cell represents a ground point, thus separating ground from non-ground areas. For example, a straight line Y = kd + b is fitted for each segment, where k represents the ground slope and b is the intercept. If the ground has a slope, the fitted line is piecewise, with different coefficients (k, b) representing different ground slopes.

[0074] S224. Select a target grid whose overall height data is lower than the ground line, and determine all point cloud data in the target grid as ground point cloud data.

[0075] For all grid cells, if the overall height data is less than or equal to the threshold, the grid cell is determined as the target grid cell, and the point cloud data in it is ground points; if the overall height data is greater than the threshold, the point cloud data in it is determined as obstacle points.

[0076] like Figure 6 As shown, the above method can set the target grid in the 2.5D grid map to gray, indicating that it is drivable. Other grids are set to black, indicating that they are not drivable, forming a 2D occupancy grid map. The area containing the gray grid is designated as the drivable area for path planning. For example, the 2D occupancy grid map is passed to the robot controller, which calculates the optimal path based on the drivable areas in the 2D occupancy grid map, avoiding obstacles and achieving dynamic obstacle avoidance. During this process, the controller updates the map data in real time to adapt to changes in the environment, ensuring the robot can move safely.

[0077] In this embodiment, to save computing power and resources, a grid map is used to represent the drivable area in the obstacle avoidance function, providing a simple and fast debugging environment for the entire vehicle boarding and alighting function link.

[0078] S230. Within the drivable area, control the robot body to move toward the door.

[0079] S240, Acquire the first depth image in the direction of the car door.

[0080] As the robot moves toward the car door, a depth camera captures a first depth image in real time in the direction of the car door to detect the position of the lower edge of the car door.

[0081] S250. Perform multi-scale sampling on the first depth image to obtain images to be detected at various resolutions.

[0082] Since the lower edge of a car door is generally a straight line, this embodiment employs an effective image line detection algorithm. Because the traditional Hough Transform algorithm suffers from drawbacks such as high computational cost, high memory consumption, sensitivity to noise, difficulty in handling complex shapes, and difficulty in parameter adjustment, this embodiment proposes a multi-scale Hough Transform algorithm to detect edge lines in the first depth image.

[0083] Optionally, the first depth image can be continuously downsampled. After each downsampling, the image resolution is reduced by a set percentage in both the row and column directions, resulting in multiple images of the target image with progressively lower resolution. This set percentage can be adjusted as needed, for example, 50%. For instance, an image pyramid algorithm can be used to continuously downsample the first depth image. An image pyramid is a multi-scale representation of an image, an effective yet conceptually simple structure for interpreting images at multiple resolutions. Each level of the image is a downsampled version of the previous level. The formula can be expressed as:

[0084] ;

[0085] in, This is the image at scale s, and Q is the downsampling filter. It is an image from the previous scale.

[0086] S260. Perform line detection on the image to be detected at each resolution to obtain edge lines of multiple resolutions.

[0087] In the image to be detected at each scale *s*, edge detection is first performed, using operators such as Canny or Sobel. After edge detection, a binary image is obtained, where edge pixels have a value of 1 and non-edge pixels have a value of 0. The Hough transform algorithm is then used to accumulate the values ​​of each detected edge pixel in the parameter space. For line detection, the parameter space is typically represented by polar coordinates, including two parameters: polar radius (distance) and polar angle (angle). For each edge pixel, based on its position in the image to be detected, the corresponding line in the parameter space is calculated and accumulated in the parameter space.

[0088] The system searches for points with the largest accumulated values ​​in the parameter space; these points represent the parameters of potentially existing lines. The points found in the parameter space are then decoded back into the image space to obtain the parameters of the detected lines. This allows the generation of edge lines for the detected image at various resolutions, i.e., multi-scale edge lines. It is assumed that the edge line parameters include slope. and intercept .

[0089] S270. Based on edge lines of various resolutions, obtain the position of the lower edge of the car door in the first depth image.

[0090] Suppose there are f resolutions, i.e. f images to be detected. If one line is detected in each image, then a total of f lines are detected. The slopes and intercepts of the f lines are averaged to obtain a final line, which is used to represent the position of the lower edge of the car door.

[0091] S280. Based on the detected position of the lower edge of the car door, control the robot body to get off the car.

[0092] In a practical application scenario, the robot needs to rely on its balance and accurate recognition of the lower edge of the car door during the disembarkation process to avoid falls or collisions, ensuring the safety and efficiency of the process. Taking a robot dog as an example, a robot dog is a robot capable of simulating the movements and behaviors of a real dog, possessing autonomous movement and self-stabilization capabilities. Robot dogs are equipped with various sensors, including but not limited to gyroscopes, accelerometers, and tilt sensors, which monitor the robot dog's posture and movement status in real time. Through these sensors, the robot dog can perceive its tilt angle and direction of movement, thus making corresponding adjustments. When the robot dog detects that it is tilting or has lost its balance, the control algorithm reacts immediately, adjusting the robot dog's joints and movement trajectory to maintain balance. Besides intelligent perception and control algorithms, the robot dog's structural design also plays a crucial role in its self-stabilization. Robot dogs can be wheeled or legged (see appendix). Figure 7 The design provides better stability and flexibility.

[0093] Once the robot reaches the ground, it can again use its depth camera and LiDAR to identify the designated location, accurately locate and navigate to it. Furthermore, the robot can activate its autonomous charging function upon reaching the ground, allowing it to automatically navigate to a charging station for recharging and prepare for its next mission.

[0094] In this embodiment, the point cloud data is first rasterized, converting the 3D point cloud data into a 2D occupancy raster map. This process, by mapping the point cloud data onto the raster, significantly improves processing efficiency, enabling the rapid generation of a 2D occupancy raster map, which can be directly used for path planning and control. To address slopes and irregular terrain, this solution proposes an obstacle avoidance algorithm based on ground segmentation. This algorithm, by segmenting the ground, can adapt to different terrain variations, thereby enhancing the robot's ability to move in various scenarios. The multi-scale Hough transform in this embodiment is an extension of the Hough transform, capable of detecting geometric shapes of different sizes; by applying the Hough transform at multiple scales, the detection flexibility and applicability are improved, thereby enhancing the accuracy of edge detection.

[0095] In summary, this application significantly improves the overall efficiency and adaptability of the robot through optimizations in point cloud rasterization processing, multi-scale Hough transform algorithm, and autonomous disembarkation function.

[0096] This application also provides a robot, see [link to relevant documentation] Figure 7The system includes a controller (built into the robot body) and the robot body connected to the controller. This embodiment does not limit the robot's shape; for example, it may resemble a robotic dog. The robot body includes at least a forward drive mechanism and a steering mechanism, allowing it to move forward or turn within the vehicle under the control of the controller. A sensing device is installed on the robot body to acquire images in the direction of the vehicle door. In response to a disembarkation command, the controller determines a drivable area from its current position to the vehicle door; within this drivable area, it controls the robot body to move towards the vehicle door; it acquires a first depth image in the direction of the vehicle door and performs door edge detection on the image to obtain the position of the lower edge of the vehicle door; based on the detected position of the lower edge of the vehicle door, it controls the robot body to disembark.

[0097] The controller is also used to execute the method for autonomous robot disembarking provided in any of the above embodiments, and has the corresponding technical effects, which will not be described in detail here.

[0098] Optional, see below Figure 7 The sensing devices include a LiDAR scanner and a depth camera. The LiDAR scanner collects point cloud data from the current position to the door position; the depth camera, in response to a disembarkation command, collects a second depth image of the robot's surroundings; a first depth image of the robot's surroundings; and a third depth image of the area outside the door. The robot is equipped with a battery to charge the sensing devices, forward drive mechanism, and steering mechanism.

[0099] Optionally, the robot is also equipped with a standard camera for pedestrian tracking and facial recognition. To ensure that most human features can be seen at close range, both the standard camera and the LiDAR are tilted upwards at a certain angle. Real-world testing showed that the standard camera and LiDAR have virtually no blind spots at close range.

[0100] Optionally, the robot is also equipped with a microphone and a voice processing device to collect and semantically analyze the vehicle's disembarkation commands.

[0101] The controller, microphone, voice processing unit, sensing unit, forward drive unit, and steering unit communicate using the ROS2 (Robot Operating System 2) mechanism. ROS2 is an open-source platform for robot development that provides a suite of tools and libraries for building robot applications. ROS2 is designed with greater emphasis on real-time performance, safety, and reliability, making it suitable for a wider range of applications and hardware platforms.

[0102] This application also provides a robot autonomous disembarkation system, including a robot and a vehicle. The robot is located inside the vehicle. The vehicle is used to issue disembarkation commands to the robot; after issuing the disembarkation command, it detects whether the vehicle door is open; if it is not open, it notifies personnel to open the vehicle door, or controls the vehicle door to open automatically.

[0103] In summary, this application utilizes the robot's own LiDAR and depth camera for efficient environmental perception and self-localization through a sensing device, avoiding dependence on external signals. Point cloud rasterization processing enhances the system's flexibility and processing efficiency, while an innovative ground segmentation algorithm adapts to various terrains. The autonomous dismounting function, combining drivable area and edge detection, improves dismounting accuracy and safety. These technologies collectively enhance the robot's overall efficiency and adaptability in complex environments.

[0104] The above description is merely a preferred embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for a robot to autonomously disembark, characterized in that, include: In response to the disembarkation command, determine the drivable area from the current location to the door location; Within the drivable area, the robot body is controlled to move toward the vehicle door; Acquire a first depth image in the direction of the car door, and perform door edge detection on the first depth image to obtain the position of the lower edge of the car door; Based on the detected position of the lower edge of the car door, the robot body is controlled to disembark, including: acquiring a third depth image outside the car door, and performing edge detection on the disembarkation platform to obtain the position of the disembarkation platform edge; combining the drivable area with the position of the disembarkation platform edge to obtain a safety map, specifically marking the disembarkation platform edge on the boundary of the drivable area to obtain a safety map; and controlling the robot body to disembark along the disembarkation platform within the area corresponding to the safety map.

2. The method according to claim 1, characterized in that, Perform door edge detection on the first depth image to obtain the position of the lower edge of the door, including: Multi-scale sampling is performed on the first depth image to obtain images to be detected at multiple resolutions; Line detection is performed on the image to be detected at each resolution to obtain edge lines at multiple resolutions; The position of the lower edge of the car door in the first depth image is obtained based on the edge lines of the various resolutions.

3. The method according to claim 2, characterized in that, Multi-scale sampling is performed on the first depth image to obtain detection images of various resolutions, including: The first depth image is continuously downsampled. After each downsampling, the image resolution is reduced by a set ratio in the row and column directions to obtain multiple images to be detected with gradually decreasing resolution.

4. The method according to claim 1, characterized in that, The process of determining the drivable area from the current location to the door location in response to an exit command includes: In response to the disembarkation command, a second depth image of the area around the robot body is acquired; Based on the depth data in the second depth image, determine the direction and position of the door; The robot body is controlled to face the car door, and the drivable area from the current position to the car door position is determined.

5. The method according to claim 1, characterized in that, Determining the drivable area from the current position to the door position includes: Collect point cloud data from the current position to the door position; The point cloud data is segmented into ground data to obtain ground point cloud data, which constitutes the drivable area.

6. The method according to claim 5, characterized in that, The point cloud data is segmented into ground data to obtain ground point cloud data, and the method further includes: The point cloud data is rasterized to obtain a raster map, and each grid in the raster map includes joint height data of multiple point cloud data. Each grid cell in the grid map is mapped to multiple sector regions, each sector region includes multiple containers, and each container is mapped with the combined height data of multiple grid cells; The minimum integrated height information of each container is determined in each sector region, and piecewise linear fitting is performed on the minimum integrated height information of each container in each sector region to obtain the ground line; Select a target grid whose overall height data is lower than the ground line, and determine the point cloud data in the target grid as ground point cloud data.

7. The method according to claim 1, characterized in that, Based on the detected position of the lower edge of the car door, the robot body is controlled to disembark, including: The height of the lower edge of the car door from the ground is detected; if the height is less than or equal to the maximum jump height of the robot body, the robot body is controlled to jump off the car from the lower edge of the car door.

8. A robot, characterized in that, include: The controller and the robot body connected to the controller; The robot body is equipped with a sensing device for collecting images of the direction of the car door; The controller is used to respond to a disembarkation command by determining a drivable area from the current position to the door position; within the drivable area, controlling the robot body to move towards the door; acquiring a first depth image in the direction of the door, and performing door edge detection on the image to obtain the position of the lower edge of the door; based on the detected position of the lower edge of the door, controlling the robot body to disembark, including: acquiring a third depth image outside the door, and performing disembarkation panel edge detection on the third depth image to obtain the position of the disembarkation panel edge; combining the drivable area with the position of the disembarkation panel edge to obtain a safety map, specifically marking the disembarkation panel edge on the boundary of the drivable area to obtain a safety map; within the area corresponding to the safety map, controlling the robot body to disembark along the disembarkation panel.

9. The robot according to claim 8, characterized in that, The sensing device includes: LiDAR is used to collect point cloud data from the current position to the position of the car door; A depth camera is used to acquire a second depth image around the robot body in response to a disembarkation command; a first depth image around the robot body; and a third depth image outside the door.

10. A robot autonomous unloading system, characterized in that, include: The robot and vehicle as described in claim 8 or 9; The robot is located inside the vehicle; The vehicle is used to issue a disembarkation command to the robot.

Images