WO2012086029A1 - Autonomous movement system - Google Patents

Abstract

The autonomous movement system of the present invention for traveling to a destination while also estimating a self-position by referring to a map on the basis of measurement data from a peripheral object shape measurement unit is characterized in comprising: a travel environment shape model input unit for inputting a shape model of the travel environment; a self-position error prediction unit for calculating the prediction error of the self-position estimation on the basis of the shape model; a prediction error map generation storage unit for generating and storing, in a region where the autonomous movement system is capable of traveling, a prediction error map with which the prediction error is associated; and a path planning unit for referring to the prediction error map and planning a path on the basis of the prediction error corresponding to the path in which the autonomous movement system is traveling.

Description

Autonomous mobile system

The present invention relates to an autonomous mobile system that moves an autonomous mobile body to a destination while estimating a self-position based on data of a measuring device.

Conventionally, based on the data of a measurement device (external sensor or internal sensor) mounted on a vehicle, it self-moves in the real environment according to a planned route while estimating its own position on a map corresponding to the real environment. Thus, autonomous mobile systems that reach their destinations are known (see, for example, Patent Documents 1 to 3).

JP 2004-110802 A JP 2004-219332 A JP 2008-165275 A

The autonomous mobile system in the prior art is provided with a map in which object shapes in a real environment that can be used as landmarks (marks) used for estimating the self-location are stored. Self-position estimation has been realized by matching (positioning) this map with the shape of surrounding objects existing in the actual environment measured using a measuring device such as a laser scanner.

However, since the measurement range of a measurement device such as a laser scanner is limited, it is not possible to detect a landmark sufficient for accurately estimating the self position depending on the moving path. Alternatively, only landmark shapes that have geometrically large matching errors (displacements) can be measured, and between the self-position estimated by the autonomous mobile system and the actual position where the autonomous mobile system is actually located. There was a case where a big error was born.

As a result, even when the autonomous mobile system itself recognizes that it is moving along the target route, the autonomous mobile system may actually deviate from the route. As a result, the autonomous mobile system collides with obstacles (static obstacles such as buildings, roadside trees, and power poles, and dynamic obstacles such as people and other vehicles that are moving objects) or breaks at intersections. There was a possibility of going on the wrong path. In addition, there is a problem that the autonomous mobile system completely loses its own position and it is difficult to continue the subsequent movement.

An object of the present invention is to provide an autonomous mobile system that plans a route that allows the autonomous mobile system to move appropriately without losing sight of its own position and reaches the destination.

The above-mentioned purpose is to input a travel environment shape model in an autonomous mobile system in which an autonomous mobile body travels to a destination while estimating its own position by referring to a map based on measurement data from a peripheral object shape measurement unit. An environmental shape model input unit, a self-position error prediction unit that calculates a prediction error of self-position estimation based on information from the driving environment shape model input unit, and the prediction error in an area where the autonomous mobile body can travel A prediction error map generation storage unit that generates and stores a predicted error map, and a route plan unit that performs route planning based on the prediction error corresponding to the route traveled by the autonomous mobile body with reference to the prediction error map This is achieved by providing

Further, the object is that the travel environment shape model input unit, the self-position error prediction unit, and the prediction error map generation storage unit are installed in a management system, the route planning unit is installed in a terminal system, and the management system And the terminal system are preferably connected by a wireless network.

Further, the object is that the prediction error map generated and stored by the prediction error map generation storage unit divides the travelable area into a plurality of small areas and holds the prediction error corresponding to the small areas. Is preferred.

Also, it is preferable that the above object holds the prediction error corresponding to each of a plurality of directions in the small area.

Further, the object is that the prediction error map generation storage unit extracts a shape having a predetermined height from the shape model input to the driving environment shape model input unit, and the self-position error prediction unit generates the prediction error map. It is preferable to calculate the prediction error using the shape extracted by the storage unit.

According to the present invention, it is possible to provide an autonomous mobile system that reaches a destination by planning a route that allows the autonomous mobile system to move appropriately without losing sight of its own position.

It is a block diagram of the autonomous mobile system which concerns on one Example of this invention. It is a block diagram of another form of the autonomous mobile system which concerns on one Example. It is the top view which extracted the driving | running | working area | region from the environmental shape model which concerns on one Example, and divided | segmented with the grid. It is a top view which shows the result of having constructed | assembled the graph structure by the node which separated the driveable area which concerns on one Example by a grid, and a link. It is a top view of the environmental shape model which has information, such as road structures, such as a building, a roadside tree, and a telephone pole, and a runnable field (road) concerning one example. It is a figure which shows the example which concerns on one Example and extracts landmark data from the measurement data of the area | region higher than a person. It is a flowchart which shows the process sequence of the self-position error estimation part which concerns on a present Example. It is a top view of the measurement range and environmental shape model which extracted the landmark data in a certain node concerning a present Example. FIG. 6 is a plan view of landmark data that has been coordinate-transformed and an environmental shape model according to the present embodiment. It is a top view of the measurement range and environmental shape model which extracted the landmark data in a certain node concerning a present Example. FIG. 6 is a plan view of landmark data that has been coordinate-transformed and an environmental shape model according to the present embodiment.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.

First, the configuration of an autonomous mobile system according to an embodiment of the present invention will be described with reference to FIG.
FIG. 1 is a block diagram of an autonomous mobile system according to an embodiment of the present invention.
In FIG. 1, an autonomous mobile system 1 is mounted on a vehicle v such as an autonomous mobile robot or an autonomous mobile vehicle and provides an autonomous mobile function. The autonomous mobile system 1 includes a traveling environment shape model input unit 2, a prediction error map generation storage unit 3, a self-position error prediction unit 4, a landmark map storage unit 5, a peripheral object shape measurement unit 6, a self-position estimation unit 7, The destination input unit 8, the route planning unit 9, and the traveling unit 10 are configured.

The individual components described above will be described. The traveling environment shape model input unit 2 is a portion for inputting a shape model of a traveling environment such as a building in a city, a road structure such as a roadside tree, and a power pole. The environmental shape model m to be input is used in recent car navigation systems, or uses CityGML (http://www.citygml.org/), which is standardized worldwide by the Open Geospatial Consortium (OGC). . These environmental shape models m have information such as the shape and texture of road structures represented by polygons and the like, and information such as sidewalks / roadways as attribute information.

The prediction error map generation storage unit 3 is a part that generates and stores a prediction error map in the travelable region using the environment shape model m input from the travel environment shape model input unit 2. When the prediction error map is generated, processing is performed in cooperation with the self-position error prediction unit 4 (specific processing contents will be described later).

The self-position error prediction unit 4 is a part that predicts in advance a self-position estimation error that occurs at each position in the travelable area on the map by using a geometric calculation method. Data necessary for the calculation is acquired from the prediction error map generation storage unit 3, and the calculated prediction error of the self-position estimation is returned to the prediction error map generation storage unit 3 (specific processing contents will be described later).

The landmark map storage unit 5 is a part that extracts a portion that can be used as a landmark (mark) from the environmental shape model m input from the traveling environment shape model input unit 2 and stores it as a landmark map. If a landmark map can be acquired separately in advance, the separately acquired landmark map may be stored without extracting the landmark from the environmental shape model m.

The peripheral object shape measuring unit 6 is a part that measures the shape of an object (a building, a road tree such as a tree, a telephone pole, a person, or another vehicle) existing around the vehicle v. For example, a laser scanner, a stereo camera, a “Time of Flight”, a Distance image camera, or the like can be used.

The self-position estimation unit 7 detects a portion to be a landmark (mark) from the shape of the surrounding object measured by the surrounding object shape measurement unit 6 at the current position, and the landmark map stored in the landmark map storage unit 5. This is a part for estimating the current self-position on the map by performing matching (position alignment).
For example, the method described in the book “Probabilistic Robotics” (author: Sebastian Thrun, Wolfram Burgard, and Dieter Fox, publisher: The MIT Press, publication year: 2005) can be used.

The destination input unit 8 is a part where the user of the vehicle v inputs a destination to the autonomous mobile system 1. For example, as in the car navigation system, the user inputs a destination by operating a touch panel displaying a map.

The route planning unit 9 is a part that plans a route from the current self-position estimated by the self-position estimation unit 7 to the destination input from the destination input unit 8 using the prediction error of the self-position estimation as a cost. . At this time, the prediction error map stored in the prediction error map generation storage unit 3 is used (specific processing contents will be described later).

The traveling unit 10 is a part for driving and driving the wheels of the vehicle v on which the autonomous mobile system 1 is mounted. Control is performed so that the vehicle v travels according to the route planned by the route planning unit 9. In addition, the movement form by a crawler and a leg may be sufficient instead of a wheel.

Next, another configuration of the autonomous mobile system 1 will be described with reference to FIG.
FIG. 2 is a block diagram of another form of the autonomous mobile system according to one embodiment.
In FIG. 2, the traveling environment shape model input unit 2, the prediction error map generation storage unit 3, the self-position error prediction unit 4, and the landmark map storage unit 5 in the autonomous mobile system 1 are not on the vehicle v, You may install in the management system 1a connected by the wireless network. A terminal system 1b is installed in the vehicle v, and the autonomous mobile system 1 is configured by combining the management system 1a and the terminal system 1b. In this case as well, each component of the autonomous mobile system 1 is the same as that of the embodiment of FIG. 1, but between the prediction error map generation storage unit 3 and the route plan unit 9, and the landmark map storage unit 5 and the self-position estimation unit 7 is different in that communication between 7 is performed via a wireless network.

Next, specific processing contents of the prediction error map generation storage unit 3 will be described with reference to FIGS.
FIG. 3 is a plan view in which a travelable area is extracted from an environmental shape model according to an embodiment and divided by a grid.
In FIG. 3, first, a travelable region (road) is extracted from the environment shape model m input from the travel environment shape model input unit 2. The top view of FIG. 3 shows a plan view of the result of extracting the travelable area from the environmental shape model m. The environmental shape model m not only has a shape and a texture, but also has information such as a sidewalk / roadway as attribute information. For example, when an autonomous mobile robot traveling on a sidewalk is targeted, By extracting, it is possible to extract the travelable area.

Subsequently, the travelable area is divided by a grid 11 or the like to express discretely. In the lower part of FIG. 3, an example is shown in which the grid 11 is divided by a uniform size. Actually, for example, the travelable area is divided by a grid 11 of 1 m square. In addition, how to divide the travelable area is not limited to the grid 11 of uniform size, but the book “Introduction to Intelligent Robots—Solution of Motion Planning Problems” (authors: Junjun Ohta, Daisuke Kurabayashi, Tomio Arai, Publishing) (A company: Corona, publication year: 2001).

FIG. 4 is a plan view showing a result of constructing a graph structure with nodes and links by dividing a travelable area according to an embodiment by a grid.
In FIG. 4, a node 12 is generated at the center of each divided grid 11. However, one round 360 deg is divided into one grid, and a plurality of nodes 12 are generated. For example, one round 360 deg is divided into eight every 45 deg to generate eight nodes 12. Therefore, each node 12 becomes a node 12 in a three-dimensional space of (x, y, θ). By providing the information on the azimuth θ, it is possible to consider that the measurement range 14 of the peripheral object shape measurement unit 6 cannot cover the entire circumference, and the measurable region changes depending on the azimuth θ. Except for some peripheral object shape measuring units 6 such as an omnidirectional stereo camera, many peripheral object shape measuring units 6 such as a laser scanner cannot cover the entire circumference, so each node 12 has information on the direction θ. It is effective to have

Furthermore, a graph structure is constructed by connecting nodes corresponding to 8 neighbors as viewed in the two-dimensional plane of (x, y) with links 13. This graph structure is the basic data structure used by the route planning unit 9. However, there is an embodiment of the route planning unit 9 that does not use the link 13 (details will be described in the description of the route planning unit 9).

Subsequently, the prediction error map generation storage unit 3 generates data for passing to the self-position error prediction unit 4. The self-position error prediction unit 4 generates data necessary for calculation for each node 12 in order to predict an error of self-position estimation for each node 12.

FIG. 5 is a plan view of an environmental shape model having information on road structures such as buildings, roadside trees, utility poles, and travelable areas (roads) according to an embodiment.
FIG. 6 is a diagram illustrating an example of extracting landmark data from measurement data in a region higher than a person according to the present embodiment. Hereinafter, the node 12a in FIG. 5 will be described as an example.
In FIG. 5, first, measurement data is generated by simulating an object that exists at the position (x, y, θ) of the node 12 a and exists in the measurement range 14 of the peripheral object shape measurement unit 6. For example, when a laser scanner is used as the peripheral object shape measurement unit 6, the measurement data can be simulated by performing ray casting, which is a computer graphics technique, and obtaining the reflection point of the laser irradiated from the laser scanner. It is.
Note that the measurement data simulated at this time may be thinned out evenly. For example, thinning is performed at intervals of 0.1 m.

Subsequently, landmark data z used as a landmark is extracted from the measured measurement data. For example, as shown in FIG. 6, in the case where measurement data of a region d higher than the person h is used as a landmark, the landmark data is extracted by extracting a predetermined height using the height information of each measurement data. Extract z. Alternatively, data used as a landmark may be extracted from the result of clustering measurement data or extracting features.

The above processing is performed for each node 12 and the landmark data z is extracted for each node 12. The landmark data z extracted here and the environment shape model m input from the traveling environment shape model input unit 2 are data used by the self-position error prediction unit 4 for calculation.

Next, specific processing contents of the self-position error prediction unit 4 will be described with reference to FIG.
FIG. 7 is a flowchart illustrating a processing procedure of the self-position error prediction unit according to the embodiment.
In FIG. 7, the self-position error prediction unit 4 obtains a prediction error of self-position estimation as a covariance matrix Σ by geometric calculation using the landmark data z and the environment shape model m for each node 12. . The basic idea is to move the position of the landmark data z by a small amount and evaluate the degree of overlap with the environmental shape model m at that time to obtain the nature of the error (deviation) during matching, From there, the covariance matrix Σ is calculated.

If the degree of overlap with the environmental shape model m changes greatly when the landmark data z is moved, it means that the uniqueness of the shape of the landmark data z is high. Has the property of becoming smaller. Conversely, if the degree of overlap does not change much when the landmark data z is moved, this means that the uniqueness is low, and as a result, the matching error becomes large. Further, since the nature of how the error changes when moved, the nature of the error can be obtained not as a value but as a distribution represented by the covariance matrix Σ.
First, the movement amount (x _T , y _T , θ _T ) for moving the landmark data z is determined (S11).

The maximum value and the minimum value of the movement amount are set in advance, and the range is equally divided at regular intervals. The maximum value and the minimum value may be determined based on the error magnitude of the self-position estimation unit 7. For example, for _{x T} and _{y T} respectively 10 divided for each 0.5m range of + 2.5 m from -2.5m, for theta _T is 12 divided for each 10deg a range of + 60 deg from -60Deg. In this case, there are 1200 combinations of movement amounts of 10 × 10 × 12.

Next, the landmark data z is coordinate-converted with the movement amount (x _T , y _T , θ _T ) (S12). When the homogeneous transformation matrix representing the coordinate transformation by the movement amount (x _T , y _T , θ _T ) is T, the landmark data after the coordinate transformation is expressed as Tz.
The degree of overlap is calculated as the matching evaluation value _DT using the coordinate-converted landmark data Tz and the environmental shape model m (S13).

For this calculation, for example, Equation (1) and Equation (2) are used. *

First, the correspondence c between the point group of the landmark data Tz and the polygon vertex group of the environmental shape model m is obtained by the nearest neighbor search of Expression (1). Here, the number of point groups of the landmark data Tz is N. Subsequently, the matching evaluation value _DT is obtained by the root mean square of the distance between corresponding points in Expression (2) using the obtained correspondence c. Here, n is a normal direction unit vector of the polygon surface of the environmental shape model m. By calculating the inner product with the normal direction unit vector n, the distance between corresponding points is calculated as a point-to-face distance. Note that the distance between corresponding points may be calculated as a normal point-to-point distance without calculating the inner product with the normal direction unit vector n.

Then, it calculates the likelihood _{L T} from the matching evaluation value _{D T} (S14). The likelihood L _T here approximates the likelihood of overlapping with the environmental shape model m when the landmark data z is coordinate-transformed with the movement amount (x _T , y _T , θ _T ) by a normal distribution. Is. The calculation of likelihood L _T, for example using equation (3). Here, σ is a constant representing the nature of the error of the peripheral object shape measuring unit 6. A specific value may be determined based on the magnitude of the error of the peripheral object shape measurement unit 6.

The calculation of the likelihood L _T for the above movement amounts (x _T , y _T , θ _T ) is performed for all combinations of movement amounts (S15). In the above example, there are 1200 combinations.

For one node 12, After calculating the likelihood L _T combination number of any amount of movement, for example, to calculate the covariance matrix Σ using Equation (4) (S16). Equation (4), each amount of movement _{(x T, y T, θ} T) was weighted likelihood L _T for all the combinations of a calculation of the covariance matrix Σ by weighted least squares method. Here sigma _T is a mathematical symbol indicates a total for each amount of movement.

When the calculation of the covariance matrix Σ is performed for all the nodes 12, the processing of the self-position error prediction unit 4 is finished (S17, S18). Thereby, the prediction error of the self-position estimation in each node 12 in the travelable area is obtained as a covariance matrix Σ. Further, the obtained prediction error of the self-position estimation in each node 12 is returned to the prediction error map generation storage unit 3.

Accordingly, the prediction error map generation storage unit 3 can generate and store a prediction error map with a data structure of each node 12 in the travelable area and a prediction error of self-position estimation associated with each node 12. .

Subsequently, an example of specific processing contents of the self-position error prediction unit 4 will be described with reference to FIGS. 8 to 11.

FIG. 8 is a plan view of a measurement range and an environmental shape model obtained by extracting landmark data at a certain node according to one embodiment.
The processing from S11 to S14 in the flowchart of the self-position error prediction unit 4 shown in FIG. 7 in such a situation will be described as an example.

First, for example, assume that the movement amount (x _T , y _T , θ _T ) is determined to be (+1.0 m, +2.0 m, 0 deg) (S11).

Next, the landmark data z1 extracted from the measurement range 14 is coordinate-converted with the movement amount (S12).

FIG. 9 is a plan view of the landmark data subjected to coordinate conversion and the environmental shape model according to the present embodiment.
In FIG. 9, the landmark data Tz1 subjected to coordinate conversion and the environmental shape model m do not overlap (the deviation is large). Compared with before the coordinate conversion, the degree of overlap between the landmark data z1 and the environmental shape model m is greatly changed. If the degree of overlap with the environmental shape model m changes greatly when the landmark data z1 is moved, it means that the uniqueness of the shape of the landmark data z1 is high. It has the property that the matching error becomes smaller with respect to the direction.

Subsequently, a matching evaluation value _DT is calculated in order to quantitatively determine the degree of overlap between the coordinate-converted landmark data Tz1 and the environmental shape model m (S13). In FIG. 9, since the degree of overlap is low, the matching evaluation value _DT, which is the mean square of the distance between corresponding points, is a large value.

Moreover, to calculate the likelihood _{L T} from the matching evaluation value _{D T} (S14). Since the likelihood L _T is obtained by approximating the likelihood of overlap in the normal distribution, the likelihood L _T as the matching evaluation value D _T is large (low degree of overlap) is a small value.

FIG. 10 shows a plan view of the measurement range 14b from which the landmark data z is extracted and the environmental shape model m, as seen from above, in a certain node 12c in a situation different from that in FIG.

The processing from S11 to S14 in the flowchart of the self-position error prediction unit 4 shown in FIG. 7 in such a situation will be described as an example.
First, for example, it is assumed that the movement amount (x _T , y _T , θ _T ) is determined to be (+1.0 m, +2.0 m, 0 deg) as in the example of FIG. 9 (S11).
Next, the landmark data z2 extracted from the measurement range 14 is coordinate-converted with the movement amount (S12).

FIG. 11 is a plan view of the landmark data subjected to coordinate conversion and the environmental shape model according to one embodiment.
In FIG. 11, the landmark data Tz2 subjected to coordinate conversion and the environmental shape model m are relatively overlapped (the deviation is small). That is, it can be seen that there is a deviation in the x direction, but there appears to be no deviation in the y direction. Compared to before the coordinate conversion, the degree of overlap between the landmark data z2 and the environmental shape model m does not change much. If the degree of overlap with the environmental shape model m does not change much when the landmark data z2 is moved, it means that the uniqueness of the shape of the landmark data z2 is low. It has the property that the matching error increases with respect to the direction.

Subsequently, a matching evaluation value _DT is calculated in order to quantitatively determine the degree of overlap between the landmark data Tz2 subjected to coordinate conversion and the environmental shape model m (S13). In FIG. 11, since the degree of overlap is high, the matching evaluation value _DT, which is the mean square of the distance between corresponding points, is a small value.

Moreover, to calculate the likelihood _{L T} from the matching evaluation value _{D T} (S14). Since the likelihood L _T is obtained by approximating the likelihood of overlap in the normal distribution, the likelihood L _T as the matching evaluation value D _T is smaller (higher degree of overlap) is a large value.
Next, specific processing contents of the route planning unit 9 will be described.

The route plan unit 9 plans a route using the prediction error map stored in the prediction error map generation storage unit 3. As described above, the prediction error map has a data structure of each node 12 obtained by dividing the travelable area shown in FIG. 4 by the grid 11 and a prediction error of self-position estimation associated with each node 12. Each node 12 is a node 12 in a three-dimensional space of (x, y, θ), and there are a plurality of nodes 12 having different azimuths θ in the same place as seen in the two-dimensional plane of (x, y). .

As an embodiment of the route planning unit 9, at least two forms are conceivable. The first is a form in which a route plan is performed by a shortest route search algorithm in the graph theory for a graph structure constructed by connecting nodes with links 13 as described above. The shortest path search is performed using the prediction error of the self-position estimation of each node 12 as a cost. Second, using the prediction error of self-position estimation of each node 12 as the cost of the grid 11, the paper “Route Search Problem-Robot Motion Planning-” (Author: Hirohisa Hiragawa, Publisher: Journal of Information Processing Society of Japan, Publishing) This is a form of route planning using the global potential method described in 1994). When the global potential method is used, the link 13 need not be generated.

In any embodiment, since each node 12 exists in the travelable region, the travelable region is discretely expressed in a three-dimensional (x, y, θ) configuration space, and the target route in the region is represented. Can be requested. The current self-position estimated by the self-position estimation unit 7 is set as the start position, the destination input from the destination input unit 8 is set as the goal position, and the path planning is performed using the prediction error of the self-position estimation of each node 12 as the cost. Do.

Since the prediction error of the self-position estimation possessed by each node 12 is obtained as a distribution represented by the covariance matrix Σ, when the path planning unit 9 calculates the cost, the plurality of nodes 12 through which the path passes have. By fusing the covariance matrix Σ by maximum likelihood estimation, the covariance at the time of passing through each node 12 is obtained. In addition, simulation of covariance when traveling along a route is performed using the driving model described in the book “Probabilistic Robotics” (Author: Sebastian Thrun, Wolfram Burgard, and Dieter Fox, Publisher: The MIT Press, Publication year: 2005) Alternatively, the cost can be obtained by fusing the covariance matrix Σ of the node 12 with the maximum likelihood estimation.

By using the prediction error represented by the fused covariance as the cost of the path plan, it is determined whether the covariance at the time of passing through each node 12 exceeds a predetermined threshold, thereby increasing the prediction error. It is possible to plan a route to the destination considering that the route does not pass. To determine whether the covariance exceeds a predetermined threshold, the elements of the matrix may be compared, or the eigenvalues of the matrix may be compared.

A distribution of prediction errors in which each node 12 has a cost as a value, and each node 12 represents the cost as a covariance matrix Σ, instead of evaluating the route by the sum of the costs of a plurality of nodes 12 through which the route passes. And the route is evaluated by covariance when the route passes through each node 12. This makes it possible to plan a route in consideration of differences in the nature (size and direction of deviation) of errors that occur in self-position estimation at each position. Based on this cost evaluation method, it is described in the book “Introduction to Intelligent Robots—Solutions for Motion Planning Problems” (authors: Junjun Ota, Daisuke Kurabayashi, Tomio Arai, publisher: Corona, publication year: 2001) By using a route planning method (shortest route search or global potential method in graph theory), the route planning is performed taking into account the prediction error of self-position estimation (a route with a large prediction error should be avoided, and the destination Plan the route to

As described above, according to the present invention, by using the autonomous mobile system having the above-described configuration, an error in self-position estimation at each position in the travelable region is predicted in advance, and a route that increases the prediction error does not pass. It is possible to plan the route to the destination in consideration of the above.

Therefore, this invention can acquire the outstanding effect of implement | achieving the autonomous movement system which can move to the destination appropriately, without losing sight of a self-position.

DESCRIPTION OF SYMBOLS 1 ... Autonomous movement system, 2 ... Driving environment shape model input part, 3 ... Prediction error map production | generation storage part, 4 ... Self-position error prediction part, 5 ... Landmark map storage part, 6 ... Surrounding object shape measuring unit, 7 ... Self-position estimating unit, 8 ... Destination input unit, 9 ... Route planning unit, 10 ... Traveling unit, 11 ... Grid, 12 ... node, 13 ... link, 14 ... measurement range.

Claims (5)

In an autonomous mobile system in which an autonomous mobile body travels to a destination while estimating its own position by referring to a map based on measurement data from a surrounding object shape measurement unit,
A driving environment shape model input unit that inputs a driving environment shape model; a self-position error prediction unit that calculates a prediction error of self-position estimation based on information from the driving environment shape model input unit;
A prediction error map generation storage unit that generates and stores a prediction error map in which the prediction error is associated with the autonomous mobile body in a travelable region;
An autonomous mobile system comprising: a route planning unit that performs route planning based on the prediction error corresponding to a route traveled by the autonomous mobile body with reference to the prediction error map. In the autonomous mobile system according to claim 1,
The travel environment shape model input unit, the self-position error prediction unit, and the prediction error map generation storage unit are installed in a management system,
The route planning unit is installed in a terminal system, and the management system and the terminal system are connected by a wireless network. In the autonomous mobile system according to claim 1,
The prediction error map generated and stored by the prediction error map generation storage unit divides the travelable area into a plurality of small areas, and holds the prediction error corresponding to the small areas. Moving system. In the autonomous mobile system according to claim 3,
The autonomous mobile system, wherein the prediction error corresponding to each of a plurality of directions is held in the small area. In the autonomous mobile system according to claim 1 or 2,
The prediction error map generation storage unit extracts a shape having a predetermined height from the shape model input to the driving environment shape model input unit,
The autonomous mobile system characterized in that the self-position error prediction unit calculates the prediction error using the shape extracted by the prediction error map generation storage unit.

Images