Search

CN-114600054-B - Computer-implemented method for creating an environment map for operating a mobile agent

CN114600054BCN 114600054 BCN114600054 BCN 114600054BCN-114600054-B

Abstract

The invention relates to a computer-implemented method for operating a mobile agent (1) in an environment based on the position of the mobile agent (1), wherein for operating the mobile agent (1), the positioning of the mobile agent (1) is carried out by detecting (S1) sensor data with respect to a wall (2) and/or an object (3) located in the environment, wherein the sensor data indicate the alignment and distance of the wall (2) and/or the object (3) in an agent coordinate system (A) fixed with respect to the agent, and determining (S11-S18) the position of the mobile agent (1) in the environment by means of a SLAM algorithm, taking into account the Manhattan orientation.

Inventors

  • P. Biebel

Assignees

  • 罗伯特·博世有限公司

Dates

Publication Date
20260505
Application Date
20200924
Priority Date
20191107

Claims (15)

  1. 1. A computer-implemented method of operating a mobile agent (1) in an environment based on the pose of the mobile agent (1), wherein for operating the mobile agent (1) the positioning of the mobile agent (1) is performed with the following steps: -detecting sensor data about walls (2) and/or objects (3) located in an environment, wherein the sensor data is descriptive of an alignment and a distance of the walls (2) and/or objects (3) in an agent coordinate system (a) fixed relative to the agent; -determining the pose of the mobile agent (1) in the environment by means of a SLAM algorithm taking into account manhattan orientations, wherein for each determination of an effective manhattan orientation in a detection step a manhattan node and a manhattan edge between one of the pose nodes and the involved manhattan node are inserted into the SLAM graph, such that a subsequent optimization algorithm for minimizing errors is next used, said additional boundary conditions being predefined by manhattan orientation in order to determine the pose of each of the pose nodes, wherein the SLAM algorithm corresponds to a graph-based SLAM algorithm based on a SLAM graph with nodes and transformation edges, wherein a respective pose node describes the pose of the mobile agent determined from the sensor data, and wherein between each two pose nodes there is a transformation edge describing a change in the pose determined from the sensor data between the poses associated with the respective two pose nodes, wherein the pose nodes are subjected to a respective algorithm, wherein a respective error is determined from the manhattan node and the manhattan node, and a respective threshold error is determined from the graph, wherein a probability of the change in the manhattan node and the manhattan node is subjected to the graph is determined, and the error is taken into account.
  2. 2. The method of claim 1, wherein the SLAM algorithm sets a minimization of an error function related to a measurement error of the sensor data, which is related to the manhattan orientation.
  3. 3. Method according to claim 1, wherein a set of points is determined from the sensor data, the set of points describing the detected coordinates of the wall (2) and/or object (3), wherein the points in the set of points are combined into local units adjacent to each other, wherein manhattan normalization is applied to units having a unit orientation with a straight line or a 90 ° structure, wherein the manhattan orientation is determined from the dominant orientation of the straight line structure in the local units.
  4. 4. A method according to claim 3, wherein, to determine the dominant orientation of the cells, similarity to all further cell orientations is determined for each cell orientation of one cell, and a similarity measure is determined for each cell, wherein the manhattan orientation corresponds to a cell orientation of a selected one of the cells according to the similarity measure.
  5. 5. The method of claim 4, wherein the determining of the similarity is performed based on a manhattan normalized orientation difference between each of the two cell orientations.
  6. 6. The method of any of claims 3-5, wherein the determined manhattan orientation is considered only if more than a predetermined number of cell orientations of the cells lie within a manhattan normalized orientation difference from the determined manhattan orientation.
  7. 7. Method according to claim 1 or 2, wherein the mobile agent (1) is operated according to the determined pose, and/or wherein an environment map is determined by means of the current pose of the mobile agent (1), and the mobile agent (1) is operated according to the environment map.
  8. 8. The method of claim 4, wherein a similarity measure is determined for each cell as a sum of the similarities of the respective cell with respect to the remaining cells.
  9. 9. The method of claim 4, wherein the manhattan orientation corresponds to a cell orientation of the one of the cells having the largest similarity metric.
  10. 10. The method of claim 5, wherein the similarity is determined from an exponential function of the negative manhattan normalized difference.
  11. 11. A device for operating a mobile agent (1) based on the pose of the mobile agent (1) in an environment, wherein the device is configured to perform a localization of the mobile agent (1) according to the method according to claim 1 by: -detecting sensor data about walls (2) and/or objects (3) located in an environment, wherein the sensor data is descriptive of an alignment and a distance of the walls (2) and/or objects (3) in an agent coordinate system (a) fixed relative to the agent; -determining the pose of the mobile agent (1) in the environment by means of a SLAM algorithm taking into account the manhattan orientations, wherein for each determination of an effective manhattan orientation in the detection step a manhattan node and a manhattan edge between one of the pose nodes and the involved manhattan node are inserted into the SLAM graph, such that a subsequent optimization algorithm for minimizing errors next uses additional boundary conditions, which are predefined by the manhattan orientation, in order to determine the pose of each of the pose nodes.
  12. 12. The device according to claim 11, wherein the device is a control unit (11).
  13. 13. A mobile agent (1) having a device according to claim 11, an environment detection sensing means (12) for providing sensor data and having a movement actuation means (13), the movement actuation means (13) being configured to move the mobile agent (1).
  14. 14. A computer program product with instructions, which is set up to implement the method according to any one of claims 1 to 10 when the instructions are executed on a computing unit.
  15. 15. A machine-readable storage medium having the computer program product of claim 14 stored thereon.

Description

Computer-implemented method for creating an environment map for operating a mobile agent Technical Field The present invention relates to the field of autonomous control of mobile agents (eines mobilen Agenten) within a region of motion. In particular, the present invention relates to a method for locating mobile agents in an environment. Background Mobile agents are known for many applications. Here, the mobile agent is to move independently in an environment equipped with a boundary (wall) and an obstacle (object). Such mobile agents may include, for example, robots, vehicles, autonomous cleaning robots, autonomous mowers, and the like. The ability to determine the exact pose (position and orientation) of a moving agent in an environment defined by sensor data and to determine an environment map associated with the exact pose is the basis for trajectory planning of a path of motion within the environment. Conventional behavioral approaches for creating an environment map have provided for the use of mobile agent environment detection sensing devices, which may include, for example, lidar systems, cameras, inertial sensing devices, range sensing devices (Odometriesensorik), and the like. To create an environment map, the mobile agent is typically moved in a controlled manner or automatically and evaluates the sensor data of the environment detection sensor device in order to create an environment map. A conventional method for creating an environment map provides for the application of a so-called SLAM algorithm which processes the recorded data of the environment detection sensor device and determines therefrom the pose of the mobile agent and also the corresponding environment map. Disclosure of Invention According to the invention, a method for locating a mobile agent within an environment is provided, as well as a device for locating a mobile agent, a device for controlling a mobile agent in an environment and a mobile agent. Other construction schemes are also described below. According to a first aspect, a method for operating a mobile agent in an environment based on its pose is provided, wherein the localization of the mobile agent is performed by: -detecting sensor data about walls and/or objects located in an environment, wherein the sensor data is indicative of the alignment and distance of the walls and/or objects in A (AGENTENFESTEN) agent coordinate system fixed relative to the agent; -determining the pose of the mobile agent in the environment by means of SLAM algorithm taking into account Manhattan orientation (Manhattan-Orientierung). A fundamental disadvantage of SLAM algorithms for locating mobile agents, that is to say SLAM algorithms for determining the absolute pose of a mobile agent within an environment, is the drift accumulation of the detected sensor data. In particular, due to the coupling, detection errors with respect to the determined orientation cause a great drift of the coupled position of the mobile agent. Although the orientation drift can be compensated by a direct measurement alignment, this is usually possible only by compass measurements which are not available in many cases, in particular in applications in buildings or in environments which are strongly loaded by stray magnetic fields. The above method sets up that the SLAM algorithm is combined with the assumption of the manhattan world. The manhattan world concept is that many artificial environments have a regular structure, especially many walls are parallel or at right angles to each other, and the walls usually run straight over a long stretch of road and are flat. For example, the Manhattan world concept is known in publication B. Beasey et al, "Accurate On-Line 3D Occupancy Grids Using Manhattan World Constraints" (2012, IEEE/RSJ International Conference On Intelligent Robots AND SYSTEMS, velarmora (Vilamoura), pages 5283 to 5290). Furthermore, the SLAM algorithm may set a minimization of an error function related to the measurement error of the sensor data, the error function related to the Manhattan orientation. According to one embodiment, the SLAM algorithm may correspond to a graph-based SLAM algorithm based on a SLAM graph having nodes and edges, wherein the SLAM graph has gesture nodes that each illustrate a gesture of the mobile agent that is determined from the sensor data, and has a transformed edge between each two gesture nodes, wherein the transformed edge illustrates a gesture change that is determined from the sensor data between the gestures associated with each two gesture nodes, wherein a manhattan node and at least one manhattan edge between the manhattan node and one of the gesture nodes are added according to a manhattan orientation. In addition, the SLAM algorithm may use an error function determined from the SLAM map, which is considered by the error probabilities associated with the nodes and edges, which determines the revised pose associated with the nodes of the SLAM map. In princi