JP2011100306A - Simulation system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To perform a simulation capable of reproducing a motion nearly equivalent to the one in a real world. <P>SOLUTION: A simulation system includes: a simulation unit for simulating data that can be obtained by a sensor installed in a robot; a sensor data correction unit for determining whether points with attributes added thereto are included in obtained sensor data obtained by the sensor, performing correction according to a sensor data correction parameter on a group of points with attributes included therein, and outputting corrected sensor data; a virtual robot position and posture estimation unit for estimating a position and posture of a robot in a 2D environment map through a process of comparison between the output corrected sensor data and an attribute reflection 2D environment map, and outputting estimated virtual robot position and posture data; a movement target position determination unit for determining a subsequent movement destination; and a movement unit for updating a virtual robot position and posture according to a determined value. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

   The present invention relates to a simulation system, and more particularly to a robot simulation system that verifies the behavior of an autonomous mobile robot in a virtual space.

  A lot of robots that carry goods etc. have been introduced at sites such as factories and warehouses. In order to operate in an actual factory, it is necessary to sufficiently verify safety and the like in advance. Whether the robot movement algorithm is correct, what kind of transfer route is efficient, whether there are collisions when operating multiple robots at the same time, whether there is a possibility of position estimation error or route deviation, and causes deadlock between robots There are various verification items, such as whether or not there is a possibility.

  As a control system for a plurality of robots that move autonomously inside a factory or the like, there is an operation control system (Patent Document 1) listed below. The centralized control device grasps the robot position by communicating with each robot, and makes an operation plan of the robot closest to the place where the work is required, but by combining a predetermined short distance route, A route is created, the position of each robot is grasped by communication, and the robot can move along an optimum route without colliding with each other. The possibility of interference of each robot can be predicted by simulation.

  In addition, it is assumed that the robot moves autonomously while recognizing the surrounding environment with the sensor of the robot, and the robot position is estimated based on the virtually obtained camera image, and the operation is simulated. There is also a robot simulator (Non-patent Document 1).

JP 2005-242489 A

“4D space simulator for image understanding,” IEICE Technical Report, No.PRMU-97-277, pp.89-96, 1998.

  The simulator described in Patent Document 1 holds a map of a traveling area and confirms a position by appropriately detecting a target (such as a mark on the ground) while traveling, but it is necessary to set a target in advance. When the layout in the factory is changed, it is necessary to re-install everything.

  The simulator described in Non-Patent Document 1 performs operation verification in a simulated environment created with CG, and is likely to be in a situation different from the situation in the real space. Also, it is basically a simulator for algorithm verification, and is not a simulator that is supposed to travel in real space.

  The present invention has been made in view of the above-mentioned reasons, and its purpose is to use an algorithm used in a real machine in a virtual space that reproduces a situation close to reality without installing a mark in the environment. Another object of the present invention is to provide a simulation system that recognizes and moves the surrounding environment based on sensor data that may be actually obtained.

  A desirable mode of the present invention is the following simulation system. That is, the simulation system according to the present invention receives the attributed virtual space model that reproduces the three-dimensional shape of the space and adds the attributes, and the position and orientation of the robot in the virtual space model, and is mounted on the robot. A sensor simulation unit that simulates data that can be acquired by the sensor, and whether the acquired sensor data acquired by the sensor includes a point to which an attribute is added, and for a point group that includes the attribute Then, the sensor data correction unit that performs correction according to the sensor data correction parameter and outputs the corrected sensor data, and the output correction sensor data and the attribute reflection two-dimensional environment map are input, and the two-dimensional matching process is performed on both data. Estimate the position and orientation of the robot in the environmental map, and output virtual robot estimated position and orientation data A virtual robot position / posture estimation unit, the estimated virtual robot position and posture, a preset route data as inputs, a movement target position determination unit that determines the next destination, and a virtual robot according to the determined value And a moving unit that updates the position and orientation.

  According to the present invention, a robot simulation closer to the real world can be realized.

The figure which shows the whole flow. The functional block diagram which showed the flow of the autonomous movement when using a real machine in a real space. The functional block diagram of an autonomous movement simulator. The block diagram of attribute reflection two-dimensional environmental map creation. The figure which showed the example of the virtual space model with an attribute. The figure which showed the example of sensor data correction | amendment. The figure which showed the example of sensor data correction | amendment. The figure which showed the example of the attribute reflection two-dimensional environmental map. The figure which showed the example of the conventional simulation. The figure which showed the example of the simulation in embodiment.

  Hereinafter, embodiments will be described with reference to the drawings.

  FIG. 1 shows the overall flow of this embodiment.

  This embodiment consists of three steps.

  First, the first step is to restore the shape of the environment based on the actually measured data and create a virtual space model. In order to perform a simulation closer to reality, it is desirable to prepare an environmental model that is as faithful as possible to the real world. One method is to create a virtual space model by measuring the environment using a distance sensor or the like and integrating the measurement data. A more accurate model can be created by using a plurality of different types of sensors and selecting and integrating appropriate data. For example, in measurement using laser light, glass has a characteristic of being transmitted and a mirror surface being reflected. However, in measurement using ultrasonic waves, the surface of a glass surface or a mirror surface can also be recognized. For this reason, accurate shape data can be obtained by using ultrasonic measurement data on the glass surface or mirror surface. In addition, the other side of the door and the wall surface cannot be recognized due to the data of the door or the wall surface when measured using laser light, but by using the measurement data by X-ray, it is possible to grasp the state inside the door or the wall surface. become able to do. Thus, by integrating data using a plurality of different sensors, it is possible to create a virtual space model that accurately reproduces the shape.

  In creating a virtual space model, the model creation means is not limited to using measured data. A model may be created with reference to a CAD drawing or the like of the target space, or an environment model equivalent to the current state may be created manually while measuring with a measure.

  In addition, assuming that there is a possibility that the situation of the site will change (the position of the object will change, a new object will be placed, or the placed object will disappear), a virtual object will be added to the created virtual space model・ Can be deleted / moved. Furthermore, not only a stationary object but also a moving object can be installed in the virtual space.

  The second step is attribute addition to the virtual space. The environment is measured using a sensor and autonomous movement is performed based on the result. In order to obtain a realistic sensing result, attributes are set in advance in the virtual space model. For attribute setting, a virtual space is displayed on a display, an attribute setting range is specified using an input device such as a mouse, a pen tablet, or a keyboard, and a predetermined attribute (for example, glass) is specified for the specified range. , Mirror, light absorption color, light expansion color, diffusion material, etc.). As another method, attributes are automatically determined from the difference or color of the acquired distance data using data obtained by measuring the same location with a plurality of sensors, for example, a laser distance sensor, an ultrasonic sensor, and a camera May be.

  The third step is to perform sensor simulation and autonomous movement simulation according to the attribute. This will be described in detail after FIG.

FIG. 2 is a functional block diagram showing the flow of autonomous movement when a real machine is used in the real space.
Here, a description will be given assuming a robot that moves while recognizing the environment with a laser distance sensor. It is assumed that the sensor mounted on the robot is fixed so that the scan plane is parallel to the xy plane (floor surface). The sensor data acquisition unit 101 measures the distance from the surrounding environment with a laser distance sensor and stores it as sensor data 201.

  The robot position estimation unit 102 receives the acquired sensor data 201 and a two-dimensional environment map 202 which is a cross-sectional view of the environment at a certain height, and determines which position / It is estimated whether the sensor is facing in the direction, and is output as the robot position 203.

  The movement target position determination unit 103 determines the position and direction where the robot should proceed next from the estimated robot position 203 and the route data 204 set on the two-dimensional environment map 202. Perform robot movement control according to the decision. By repeating the operations 101 to 104, the robot can move autonomously.

  Consider reproducing the same situation on a computer.

  That is, the basic flow of the autonomous movement simulation is as follows. It is composed of a virtual space with a three-dimensional shape that simulates a real space, a virtual robot that moves in the virtual space, a two-dimensional environment map that shows the arrangement of objects in the virtual space, and the position and direction of the virtual robot. A virtual robot position is estimated by matching a distance sensor that measures the distance to the visible surrounding environment, the measurement sensor data, and a two-dimensional environment map. Still further, the position estimation algorithm is evaluated by comparing the position and direction of the virtual robot with the position and direction of the estimated virtual robot.

  The computer includes a recording unit that records each data and each program, and a calculation unit such as a CPU that reads and executes each data and each program (not shown). Each data described in this figure is recorded in the recording unit, and each processing and each flow indicated by the functional block are read by a necessary program and executed by the arithmetic unit. These hardware configurations and processing operations are the same in other drawings described below.

FIG. 3 is a functional block diagram of the autonomous mobile simulator.
In the sensor simulation unit 301, an attributed virtual space model 401 representing a three-dimensional shape of the space and added with an attribute such as a material and the position / posture 402 of the virtual robot traveling in the virtual space are input, The distance from the virtual space is measured from the sensor position and stored as sensor data 403. The sensor data 403 is represented by a point group that is an aggregate of points, and each point has a distance / angle value between the laser distance sensor and the obstacle. Furthermore, attributed information and attribute types are added to the points where the attribute-added locations are measured. That is, the information held by each point includes the distance / angle to the obstacle, the presence / absence of the attribute, and the type of the attribute if there is an attribute.

  The sensor data correction unit 302 corrects the sensor data according to the attribute when the attribute information is included in the sensor data. The correction parameter is determined by the correction parameter 404 set for each sensor. In the real world, the data that can be acquired differs depending on the sensor used and the sensing target material. Even with the same laser distance sensor, the sensitivity to color may vary depending on the sensor, and if different types of sensors are used, there are materials that can be measured with one sensor but not with another sensor. is there. For this reason, the correction sensor data 405 is created by preparing a correction model corresponding to the sensor used and applying it to the corresponding part of the acquired sensor data.

  The robot position estimation unit 303 receives the created corrected sensor data 201 and the two-dimensional environment map 406 reflecting the attribute, and the position of the robot in which direction and direction in the two-dimensional environment map 406 is determined by collating both data. Whether the robot position 407 is output. The two-dimensional environment map 406 used here is created by the procedure shown in FIG.

  FIG. 4 is a block diagram of creating an attribute reflecting two-dimensional environment map.

  The virtual robot moving unit 306 virtually moves the robot in the virtual space. For the robot movement, for example, forward / backward / left / right movement and rotation operation are performed through input means such as a keyboard, a mouse, and a joystick, and the position / posture 402 of the virtual robot is output in accordance with the operation. The flow from the sensor simulation 301 to the robot position estimation 303 is the same as described with reference to FIG. The map update unit 307 estimates the robot position 407 by performing a matching process between the correction sensor data 405 and the attribute reflection two-dimensional map 406 created so far, and based on this result, the correction sensor data 405 is stored. By integrating into the two-dimensional map 206, the map is sequentially enlarged. For the robot position estimation 303 and the map update 307, SLAM (Simultaneous Localization And Mapping) technology that executes these two calculations simultaneously is used.

FIG. 5 is a diagram illustrating an example of a virtual space model with attributes.
The shape reproduces exactly the same situation as in the real world. The acquired data differs from the actual shape depending on the sensor characteristics of the sensor used. Therefore, an area where data different from the actual shape may be acquired is set, and an attribute is set for the area. As attributes, mirror, glass, ground glass, uneven material, light absorbing material, color, temperature, magnetism, water, etc. can be considered.

6 and 7 are diagrams showing examples of sensor data correction.
An example in which the correction parameter 404 of the laser distance sensor is applied will be described.
The sensor data obtained by the simulation is represented by a point group representing the position of the obstacle, and includes the distance and angle from the sensor position to the obstacle, and attribute presence / absence type information.

  (1-1) is an example of sensor data including a point with a mirror attribute. A point 501 represents data without an attribute, and a point 502 is data with a mirror attribute. Since the laser beam is reflected by the mirror surface, the “distance from the sensor position to the mirror + the distance to the reflected obstacle” is actually measured data. Therefore, when creating sensor data reflecting the attribute, the portion without the attribute is left as it is, and the portion with the attribute is corrected according to the distance to the obstacle at the reflection destination. For example, correction data as shown in (1-2) is created.

  (2-1) is an example of data including a point with a glass attribute. Since the laser light passes through the glass, the distance to the obstacle at the tip of the glass is added, and for example, correction data as shown in (2-2) is created.

  (3-1) is an example of data including a point with the material 1 attribute. As the sensor characteristics, an example is given in which when the material 1 is measured, it is measured farther than the actual distance. (3-2) is an example of correction data.

  (4-1) is an example of data including a point with the material 2 attribute. As the sensor characteristics, an example is given in which when the material 2 is measured, the measurement is performed with irregularities added from the actual surface. (4-2) is an example of correction data.

  (5-1) is an example of data including a point with a moving object attribute. The next time it is measured, it is likely not at that location. The probability information that there is an obstacle there is added to the point. (5-2) is an example of correction data.

FIG. 8 is a diagram showing an example of the attribute reflection two-dimensional environment map.
A virtual space two-dimensional map that accurately represents the surface shape in the space is denoted by 510. An example in which the attribute information assigned to the three-dimensional virtual space model is superimposed on the virtual space two-dimensional map is 511. In creating the environmental map by the procedure described in FIG. 4, the data acquired by the sensor simulation is added to (2-1) (2-2) in FIG. 6, (3-1) and (3- As illustrated in 2), the attribute reflecting two-dimensional map 512 is created by performing correction according to the attribute and sequentially integrating the corrected sensor data after correction. Those with the glass attribute disappear from the map, and those with the material 1 attribute remain in a reduced state.

  The map shown by 510 shows a shape faithful to the real world, and the map shown by 512 shows a shape faithful to data obtained by measuring the real world with a sensor.

  FIG. 9 is a diagram showing an example of a conventional simulation.

  The distance from the sensor to the surrounding obstacle is measured in the virtual space, and sensor data is acquired. The acquired sensor data is as shown in 403. In the simulation, the self-position is estimated by performing a matching process between the two-dimensional environment map and the acquired sensor data 403. Because the sensor data acquired in the 3D virtual space data that accurately reproduces the shape and the 2D map that is a cross-sectional view of the 3D virtual space that accurately reproduces the shape, self-position estimation can be performed correctly. it can. Therefore, it is effective for verifying algorithms such as robot position estimation. However, since sensor data different from the real world can be obtained, it is not suitable for verifying, for example, the possibility of a position estimation error, the possibility of a route deviation, and the route validity.

  FIG. 10 is a diagram showing an example of simulation in the present embodiment.

  The sensor data is acquired by measuring the distance from the sensor to the surrounding obstacle in the virtual space reflecting the attribute. The acquired sensor data is as shown in 403. Reference numeral 501 denotes sensor data without an attribute, and reference numeral 502 denotes sensor data with an attribute. The presence or absence of an attribute and the type of attribute are given to each point. In the acquired sensor data 403, a correction is performed on a point including the attribute according to the type of the attribute to generate corrected sensor data 405. In the simulation, the self-position is estimated by performing a matching process between the two-dimensional environment map 512 reflecting the attribute and the correction sensor data 405. Uses sensor data acquired in 3D virtual space data that accurately reproduces the shape and adds attributes to necessary parts, and a 2D map created by reflecting those attributes, so that self-location estimation is performed in the real world. You can get the same results as you did in. For this reason, it is effective for verification of an algorithm such as robot position estimation, and it is possible to simulate the behavior of the robot in accordance with various environmental conditions and the robot state.

  For example, the influence on position estimation when the robot is tilted can be verified. When riding on a step on the floor or something placed on the floor, the robot tilts and the sensor surface is not level with the floor. Therefore, by providing a function to tilt the virtual robot, how the sensor data acquired when the robot tilts will affect when the map is generated, and how the position will be estimated when the robot moves. Can be verified.

  For example, the position of the sensor mounted on the robot can be verified. By providing a function to change the sensor mounting method, it is possible to verify the effect on map generation and position estimation when the sensor position, for example, sensor height, angle, distance from the robot center, etc., is changed. it can. Furthermore, it is possible to verify how much the position estimation accuracy is improved by providing the function of changing the number of sensors.

  For example, it is possible to verify the influence on the position estimation and the influence on the robot behavior when a plurality of moving objects (robots, people, etc.) exist in the space. It is possible to verify how the size, distance from the robot, and movement of the moving object that blocks the sensor field of view affect the position estimation of the robot, and further, the behavior of the robot when the moving object blocks the path of the robot For example, it is possible to verify whether the selection of stopping or avoiding is appropriate.

  For example, the behavior of each robot when a plurality of autonomous mobile robots move in the same environment can be verified. There is a possibility that the robots may collide, what kind of conditions are overlapped, whether the robots are likely to cause a deadlock, where is likely to happen, and what are the overlapping conditions It is possible to verify by simulation how many vehicles can be run at the same time.

  Furthermore, it is also effective for verifying the possibility of route deviation and route validity.

  It is also effective for verifying the validity of the two-dimensional environmental map. A two-dimensional environment map is created based on distance data measured using a real machine in real space. In simulation, a virtual space created by simulating a real space is prepared, distance data is acquired by a sensor mounted on a virtual robot that travels in the virtual space, and an attribute reflection distance in which attributes are reflected in the acquired distance data The position and direction of the virtual robot are estimated by collating the data with the two-dimensional environment map in the real space. The validity of the two-dimensional environment map and the position estimation algorithm are verified by comparing the robot position / direction in the virtual space with the estimated robot position / direction. A part of the two-dimensional environment map created in the real space can be modified, added, and deleted. Furthermore, by providing a means for adding a virtual moving object created based on the moving state of the object in the real space to the virtual space to be simulated, it is possible to perform a more realistic operation verification.

  In the autonomous movement simulation in this embodiment, a virtual robot can be moved using an algorithm used for a real machine, and the influence when an actual robot is introduced can be evaluated in a PC.

400 ... Virtual space model, 401 ... Virtual space model with attributes, 402 ... Virtual robot position / posture, 403 ... Sensor data, 404 ... Sensor data correction model, 405 ... Correction sensor data , 406 ... attribute reflection two-dimensional environment map, 406-a ... two-dimensional environment map, 407 ... virtual robot estimated position / posture, 408 ... route data, 501 ... attributeless point cloud data , 502 ... Attributed point cloud data

Claims (3)

A sensor that simulates data that can be obtained by a sensor mounted on the robot, using as input the attributed virtual space model that reproduces the three-dimensional shape of the space and adds attributes, and the position and orientation of the robot in the virtual space model A simulation section;
It is determined whether or not a point with an attribute is included in the acquired sensor data acquired by the sensor, and correction according to a sensor data correction parameter is performed on the point group including the attribute. A sensor data correction unit for outputting sensor data;
A virtual robot position that receives the output correction sensor data and the attribute-reflecting two-dimensional environment map as input, estimates the position and orientation of the robot in the two-dimensional environment map by collating both data, and outputs the virtual robot estimated position and posture data・ Attitude estimation unit,
The estimated virtual robot position and posture, and preset route data as inputs, and a movement target position determination unit that determines a next movement destination;
A simulation system comprising: a moving unit that updates a virtual robot position and posture according to a determined value.   2. The simulation system according to claim 1, wherein the attributed virtual space model measures a real space using one or a plurality of sensors, and reproduces a shape based on the measured values.   2. The simulation system according to claim 1, wherein the attributed virtual space model is created by setting an attribute for a region where measurement data is changed by a sensor in a virtual space model in which a shape is reproduced. .

Images