JP6773471B2 - Autonomous mobile and environmental map updater - Google Patents

Description

The techniques disclosed herein relate to autonomous mobiles. More specifically, the present invention relates to a technology for updating an environmental map stored in an autonomous mobile body.

It stores an environmental map showing the position of obstacles in the area where the moving object moves, and is equipped with an observation device that observes the distance and direction from the moving object to the obstacle. A moving body (hereinafter referred to as an autonomous moving body) has been developed, which calculates the position of a moving body using observation results, avoids obstacles from that position, and searches for a course to reach a target position.

There are various types of obstacles, some of which are fixed and some of which are moving. Moving obstacles may not be reflected in the environmental map and may need to be modified or updated.

Patent Document 1 discloses a technique of observing the environment while moving in a region and generating an environmental map by using the movement amount of the moving body and the observation data by the distance sensor. In the technique of Patent Document 1, when the change in the environment is recognized, the captured image of the portion is presented to the operator. The operator determines whether or not it is necessary to update the environmental map of the relevant part, and updates to the latest environmental map based on this determination.

In the technique of Patent Document 2, the operator sets an area where the environment is invariant and an area where the environment is variable by using an environment map. In areas where the environment is immutable, the accumulation of errors is suppressed by prohibiting updates.

JP-A-2009-169845 Japanese Unexamined Patent Publication No. 2014-89740

The technique of Patent Document 1 requires an operator's judgment as to whether or not to update the environmental map. That is, the environment map cannot be updated autonomously by the autonomous mobile body itself. Further, in the technique of Patent Document 2, the operator needs to set a region in which the environment is invariant and a region in which the environment is variable in advance. Every time the movement area changes, the operator's processing is required. In any of the technologies, it is not possible to determine whether or not to update the environmental map only by the autonomous mobile body. The present specification discloses a technique for determining whether or not the moving body itself needs to autonomously update the environmental map without requiring processing by an operator, and updating the environmental map when the update is necessary. ..

If an environmental map is available, the distance and direction from the assumed position to the obstacle can be calculated from the environmental map by assuming the position of the autonomous moving body. Below, the information on the assumed position and the distance and direction to the obstacle calculated from the environmental map is called the calculation result.
If an observation device for observing the distance and direction from the autonomous moving body to the obstacle is prepared, the information on the distance and direction from the autonomous moving body to the obstacle can be specified from the observation result. In the following, the information on the distance and direction to the obstacle observed by the observation device is referred to as the observation result.
It is possible to assume a plurality of positions and calculate the calculation result at each of the assumed positions. In this case, it can be assumed that the autonomous moving body exists at the assumed position that gives the calculation result that best matches the observation result.

However, the environmental map is not always correct due to the existence of the moving obstacles mentioned above. In the situation where the environmental map needs to be updated, the distance and directional information calculated from the environmental map to the obstacle is incorrect and does not match the distance and directional information to the obstacle observed by the observation device. In situations where the environmental map needs to be updated, there is a problem that the exact position of the autonomous mobile cannot be identified. That is, in a situation where the environmental map needs to be updated, even if the distance and direction from the autonomous moving object to the obstacle are observed by the observation device, the exact position of the autonomous moving object cannot be specified, so the environmental map is correctly corrected. Cannot be updated. The technology for identifying the location using the environmental map and the technology for updating the environmental map have inherently incompatible factors. This specification discloses a technique for achieving both.

Assuming multiple positions, using the environmental map, calculate the distance and orientation to the obstacle at each of the assumed positions (calculate the calculation result), and the distance and orientation to the obstacle observed by the observation device (observation) According to the technique of estimating the position of the autonomous moving body by comparing the result) with the calculation result, not only the position can be estimated but also the credibility of the estimated position can be calculated. For example, if the degree of agreement between the calculation result and the observation result is high only at a specific assumed position and the degree of agreement between the calculation result and the observation result is low at all the other assumed positions, it can be said that the credibility of the estimated position is high. .. On the other hand, if the degree of agreement between the calculation result and the observation result is almost uniform for all of the plurality of assumed positions, it can be estimated that the moving body is located in the vicinity of the plurality of assumed positions. It can be said that the credibility of the estimated position cannot be estimated more clearly than that.

This specification discloses a technique that employs an algorithm that mathematically calculates the credibility and updates the environment map when the credibility decreases.

The autonomous moving object disclosed in the present specification includes an environmental map storage means, an observation device, a map-dependent estimation position calculation means, a credibility calculation means, and an environmental map update means. The environmental map storage means stores an environmental map showing the positions of obstacles existing in the area where the autonomous moving body moves. The observation device observes the distance and direction from the autonomous moving body to the obstacle. The map-dependent estimated position calculation means includes the assumed position assuming that the autonomous moving object is located on the environmental map, the distance and orientation information calculated from the environmental map to the obstacle, and the obstacle observed by the observation device. The position of the autonomous moving body is estimated from the information of the distance and direction of. The credibility calculation means calculates the credibility of the estimated position. When the credibility drops below a predetermined value, the environmental map updating means updates the environmental map stored in the environmental map storage means based on the information of the distance and the direction to the obstacle observed by the observation device.

In the above-mentioned autonomous mobile body, when the credibility drops below a predetermined value, the environmental map is updated based on the information on the distance and direction to the obstacle observed by the observation device. That is, the environment map is updated when the probability that an error exceeding the permissible range occurs in the estimated position calculated by the map-dependent estimation position calculation means rises above a predetermined value. Therefore, the necessity of updating the environment map can be determined by the autonomous moving body itself without the need for processing by the operator.

Details of the techniques disclosed herein, and further improvements, will be described in detail in the embodiments and examples for carrying out the invention.

The system configuration of the mobile body is shown. The system configuration when the moving body performs the environment map update process and the like is shown. The system configuration when the moving body executes the learning data generation process is shown. An example of a moving area in which a moving body moves during learning is shown. An example of training data is shown. The system configuration when the moving body executes the credibility calculation formula generation process is shown. The graph of the credibility calculation formula is shown. The graph of the weighting coefficient which depends on the credibility is shown. The flowchart of the environment map update process performed by a moving body is shown. The figure which visually represented the contents of the environment map update processing performed by a moving body is shown.

The main features of the examples described below are listed. It should be noted that the technical elements described below are independent technical elements and exhibit technical usefulness alone or in various combinations, and are limited to the combinations described in the claims at the time of filing. It's not a thing.

(Characteristic 1) Move the area where the environmental map is incorrect to the autonomous mobile body, and find the formula that holds between the probability distribution and the credibility.
(Feature 2) The observation device observes the distance to the obstacle in correspondence with the azimuth while changing the azimuth with respect to the autonomous moving body.
(Feature 3) The observation device is mounted on an autonomous mobile body.
(Feature 4) The position of the point cloud located on the contour of the obstacle can be derived from the environmental map.
(Feature 5) The position of the point cloud located on the contour of the obstacle can be derived from the observation device.
(Characteristic 6) The environmental map updating means is a point where the observation device observes the point stored in the environmental map when the distance observed by the observation device is shorter than the distance calculated from the environmental map at the same azimuth angle. replace.

An autonomous mobile body (hereinafter, referred to as a mobile body) of the embodiment will be described. The moving body is a wheel-driven moving body having wheels on the left and right, and autonomously travels in a predetermined area. The left and right wheels are driven independently, and the moving body can change the direction of travel by changing the rotation speed of the left and right wheels. As will be described later, the moving body repeatedly executes the process of estimating the position where the moving body exists. Each process is called a step. The "current step" described below means a step being performed by the moving body. In addition, "the previous step" means "the previous step from the current step", and "the next (next) step" means "the next (next) step from the current step". Means. The moving body travels from the starting point to the target position based on the estimated position calculated for each step. The moving body estimates its own position and the direction in which it faces, but hereinafter, the position and direction of the moving body are collectively referred to as "position".

As shown in FIG. 1, the moving body 10 includes an encoder 12, a laser range finder 14 (hereinafter referred to as LRF 14), a triangulation sensor 16, and an arithmetic unit 18.

Encoders 12 are provided for each of the motor for driving the left wheel and the motor for driving the right wheel of the moving body 10, and each encoder 12 independently detects the left wheel rotation angle and the right wheel rotation angle. The left wheel rotation angle and the right wheel rotation angle detected by the pair of encoders 12 are input to the arithmetic unit 18.

The LRF 14 emits a laser beam and observes the time until the emitted laser beam is reflected by an object and returned. From the time observed by the LRF14, the distance from the LRF14 (that is, the moving body 10) to the object is observed. Further, the azimuth angle (rotation angle from the axis fixed to the moving body) for emitting the laser beam from the LRF 14 changes with time, and scans within a predetermined azimuth angle range. The LRF 14 can observe the distance to the obstacle in association with the azimuth angle with respect to the moving body 10. Hereinafter, the distance and direction from the moving body 10 to the obstacle will be referred to as "observation results". The observation result acquired by the LRF 14 is input to the arithmetic unit 18. The LRF 14 scans the emission direction of the laser beam within a predetermined angle range (for example, within an angle range of 60 ° in each of the left and right directions) with respect to the direction fixed to the moving body 10.
The triangulation sensor 16 will be described later.

The arithmetic unit 18 includes a CPU 20 that performs arithmetic processing, a RAM 22 that temporarily stores arithmetic processing data, and a ROM 24 that stores arithmetic programs and the like executed by the CPU. A part of the ROM 24 is an EEPROM, which is rewritable.
The arithmetic unit 18 executes various processes described later. When the CPU 20 executes the calculation program stored in the ROM 24, the CPU 20 performs the map-dependent estimation position calculation unit 28 shown in FIG. 2, the map-independent estimation position calculation unit 29, and the position error calculation unit 34 shown in FIG. And so on. Since the traveling control of the moving body 10 by the arithmetic unit 18 can be performed by a known method, the estimated position calculation process and the environment map update process by the arithmetic unit 18 will be described in detail here.

(Calculation processing of estimated position of map-dependent expression)
As shown in FIG. 2, the arithmetic unit 18 includes an environment map storage unit 26 and a map-dependent estimation position calculation unit (abbreviated as the first calculation unit) 28. The first calculation unit 28 repeatedly calculates the estimated position of the moving body 10 at time intervals.

The environmental map storage unit 26 stores an environmental map stored in the EEPROM of the ROM 24 and in which information indicating the position (x, y coordinates) and height (z coordinates) of the contour of an object existing in the moving area is recorded. I remember. Objects stored in the environmental map include walls, furniture, etc. that hinder movement. The environmental map will be updated as described below. If an obstacle that did not exist at the time of preparing the environment map moves and exists in the moving area, the environment map is updated to add the existence of the obstacle. Alternatively, when the obstacles existing in the moving area at the time of preparing the environment map disappear, the existence of the obstacles is deleted from the environment map.

The first calculation unit 28 repeatedly executes the Monte Carlo position identification method (Monte Carlo Localization (hereinafter referred to as MCL)) at time intervals to calculate the estimated position of the moving body 10.
In the process of calculating the estimated position depending on the map by MCL,
(1) The integrated estimated position calculated in the previous step is used. When the current step is the first step after the moving body 10 starts moving from the starting point, the above "integrated estimated position calculated in the previous step" represents the position of the starting point.
(2) The integrated estimated position calculated in the previous step is moved according to the motion model of the moving body 10, and the estimated position at the time of execution of the current step is calculated. At this stage, the estimated position is calculated according to the motion model.
(3) A particle set distributed around the estimated position in (2) is generated. The term "particle" as used herein refers to a hypothesized position in which the calculation is carried out assuming that the particle is at that position, and includes the x-coordinate, y-coordinate, and yaw angle of the moving body 10.
(4) Next, the distance to the obstacle with respect to the azimuth is calculated for each particle from the environment map acquired from the environment map storage unit 26 and the position and yaw angle of the particles. That is, the calculation result is obtained.
(5) LRF14 also observes the distance to the obstacle with respect to the azimuth. That is, the observation result is obtained.
(6) Therefore, the distribution of the distance to the obstacle with respect to the azimuth calculated in (4) above (calculation result) and the distribution of the distance to the obstacle with respect to the azimuth observed in (5) above (observation result) are shown. Contrast.
(7) If the environmental map is accurate, the calculation result of (4) and the observation result of (5) match for one specific particle, and all the other particles (4) and (5). Will be inconsistent. In this case, it can be said that the moving body 10 has a high probability of being located on the specific particle and a low probability of being located on another particle. That is, the probability that the moving body 10 is located in each particle can be calculated from the degree of agreement between the calculation result of (4) and the observation result of (5).
(8) If the probability is calculated for each particle, the probability distribution can be obtained. It can be assumed that the moving body 10 is located at a position where the probability distribution peaks.
(9) In some cases, the particle selection is inappropriate and the correct estimated position cannot be calculated. In that case, in the probability distribution, the particle density is increased in the region where the high probability is distributed, the particle density is decreased in the region where the low probability is distributed, and the processing after (4) is repeated. As a result, the distribution of particles is optimized and the correct estimated position can be calculated.

In the above, the center position of the particles is determined by (2). On the other hand, if a standard for determining the center position of the particles cannot be obtained, the calculation may be started assuming a group of particles that spread uniformly over the entire moving area. By the particle group optimization process described above, the particle distribution is optimized and the correct estimated position can be calculated.
Instead of (2) above, the estimated position obtained by adding the movement vector (obtained from the encoder 12) for the latest one step to the integrated estimated position calculated in the previous step may be used as the center position of the particle group. Good. In this case as well, the estimated position calculated in (8) above is the position estimated depending on the map.

The first calculation unit 28 performs the above processing until the actual position of the moving body 10 (hereinafter, also referred to as a true value) is included in the set of particles and the calculation result for any of the particles becomes extremely close to the observation result. repeat. The distribution of particle sets may be updated multiple times in a single step. As a result, the first calculation unit 28 calculates the estimated position of the moving body 10 in the current step. This estimated position is estimated from the environmental map.

(Learning data generation process)
The moving body 10 performs a running called a learning running before the autonomous running. Hereinafter, the learning run of the moving body 10 and the process of generating the learning data based on the data obtained during the learning run will be described. The learning data is data used in the process of generating the credibility calculation formula described later.

As shown in FIG. 3, in addition to each unit used for autonomous driving, the calculation device 18 includes a map storage unit 30 for true value calculation, a true value calculation unit 32, a position error calculation unit 34, and a learning data generation unit. 36 and a learning data storage unit 38 are provided. The environmental map storage unit 26 stores an environmental map of an area where the moving body 10 learns and travels. The area where the moving body 10 performs the learning run may be different from the area where the moving body 10 performs the autonomous running. When the moving areas are different between the learning driving and the autonomous driving, the environmental map storage unit 26 stores the environmental maps of both moving areas. In the following, in order to distinguish between the moving area during learning driving and the moving area during autonomous driving, the former is referred to as a "first area" and the latter is referred to as a "second area". Further, the environmental map of the first area is referred to as a "first environmental map", and the environmental map of the autonomous driving area is referred to as a "second environmental map".

FIG. 4 shows an example of the first region 40 in which the moving body 10 performs a learning run. As shown in FIG. 4, fences 42 and 43, which are obstacles, are arranged in a part of the first region 40. Reflectors 44 are fixed at equal intervals on the facing surfaces (inner surfaces) of the fences 42 and 43. The travel of the moving body 10 during the learning travel is controlled by a remote controller (remote controller). The broken line indicates the traveling route of the moving body 10 during the learning traveling. As shown by the broken line, the moving body 10 travels in both the area where the fences 42 and 43 are not arranged and the area between the opposite fences 42 and 43.

The first environmental map stored in the environmental map storage unit 26 is not perfect. For example, the fence 42 is stored, but the fence 43 is not stored. The learning run and learning data generation processing are carried out under an incomplete environmental map.

The map storage unit 30 for calculating the true value is provided in the ROM 24, and stores a map in which the positions of the fences 42, 43 and the reflector 44 are recorded. The environment map stored in the true value calculation map storage unit 30 is accurate and does not require updating.

The triangulation sensor 16 emits laser light in almost all directions (360 °), and detects the laser light reflected by the reflector 44 and returned, so that the distance to the reflector 44 and the reflector 44 with respect to the moving body 10 are detected. The azimuth is observed, and this information is output to the true value calculation unit 32 of the arithmetic unit 18. The triangulation sensor 16 is mounted on the moving body 10.

The true value calculation unit 32 identifies the relative position of the moving body 10 with respect to the reflector 44 from the distance and azimuth information acquired from the triangulation sensor 16. Since the position of the reflector 44 is recorded in the map acquired from the map storage unit 30 for calculating the true value, the position of the moving body 10 on the map can be obtained by mapping the relative position of the moving body 10 to the map. Can be identified. In this way, the true value calculation unit 32 calculates the actual position (true value) of the moving body 10. This process is performed by the CPU 20. The true value of the moving body 10 is output to the position error calculation unit 34.

On the other hand, the first calculation unit 28 calculates the estimated position of the moving body 10 using the incomplete first environment map, and outputs the result to the position error calculation unit 34. The estimated position calculation process and the true value calculation process are performed at the same timing. Further, the first calculation unit 28 determines the standard deviation σ _{x in} the x direction, the standard deviation σ _{y in} the y direction, and each particle from the distribution of the latest particle set (that is, the particle set when the estimated position is calculated). The maximum value w _max of the calculated probabilities is calculated, and these values (standard deviation σ _x , standard deviation σ _y , maximum probability w _max ) are output to the learning data generation unit 36.

The position error calculation unit 34 calculates the difference (position error) between the true value acquired from the true value calculation unit 32 and the estimated position acquired from the first calculation unit 28, and outputs the difference (position error) to the learning data generation unit 36. This process is performed by the CPU 20.

When the position error acquired from the position error calculation unit 34 is 15 cm or less, the learning data generation unit 36 succeeds in estimating the position within the permissible range from the true value of the estimated position calculated by the first calculation unit 28. The estimation result is classified as "OK". On the other hand, if the position error exceeds 15 cm, it is considered that the estimated position calculated by the first calculation unit 28 deviates from the true value by more than the allowable range (that is, the position estimation fails), and the estimation result is classified as "NG". To do. In the learning run, a case where the position error is classified as "NG" is intentionally created. That is, in the actual first area 40, a fence 43 not recorded in the first environment map is arranged. For this reason, when the moving body 10 travels between the fences 42 and 43, the degree of matching between the observation result acquired by the LRF 14 and the calculation result calculated for each particle becomes low, and it becomes difficult to find the matching particle. Become. This means that even if the true value is included in the particle set, the moving body 10 cannot find a particle that matches the true value as a matching particle. Therefore, when the moving body 10 travels between the fences 42 and 43, there is a high possibility that particles different from the true value are regarded as conforming particles and the particles are calculated as the estimated position. As a result, the position error between the estimated position and the true value becomes large, and if the position error exceeds 15 cm, it is classified as "NG".

In this way, the learning data generation unit 36 classifies the estimation results depending on the map, based on the position error, that is, the amount of deviation of the estimated position of the moving body 10 with respect to the "true value". In this embodiment, the reference value of the allowable range in which the position estimation result is OK is set to 15 cm, but the reference value is not limited to this. The lower the reference value of the permissible range, the higher the prediction accuracy by the credibility calculation formula.

FIG. 5 shows an example of learning data generated by the learning data generation unit 36. As shown in FIG. 5, the learning data generation unit 36 classifies the calculation time of the estimated position depending on the map, the position error acquired from the position error calculation unit 34, the calculation result of the estimated position, and the first. Generates a data set group (training data) in which data sets associated with three variables (standard deviation σ _x , standard deviation σ _y , maximum probability w _max ) acquired from the calculation unit 28 are arranged in chronological order (note that σ). _The alphabets shown in the columns _x , σ _y , and w _max are actually numerical values). By referring to the training data, it is possible to know the position error at each time, the success / failure of the calculation result of the estimated position, and the variables when the estimated position is calculated. For example, at time t = 2s, the position error between the estimated position and the true value is 7 cm, so the estimated position calculation result is classified as "OK", and the variables σ _x , σ _y , and w _{max at} that time are , B, g, and l, respectively. Further, at time t = 10s, the position error between the estimated position and the true value is 20 cm, so the estimated position calculation result is classified as "NG", and the variables σ _x , σ _y , and w _{max at} that time are , C, h, and m, respectively. The learning data generation process is performed by the CPU 20.

The learning data storage unit 38 is provided in the ROM 24, and stores the learning data generated by the learning data generation unit 36. The learning data stored in the learning data storage unit 38 is output to the credibility calculation formula generation unit 46, which will be described later. In this way, by storing the learning data by the moving body 10 itself, it is possible to analyze the log of the learning data and verify whether or not there is an error when a problem occurs in the moving body 10. Become. Therefore, when a problem occurs in the moving body 10, it becomes easier to find the cause, and the reliability of the moving body 10 can be improved. In this embodiment, the data set includes the calculation time of the estimated position, but the configuration is not limited to this. For example, the data set may not include the time, and the training data may be composed of a group of data in which the data set is arranged independently of the time.

(Credibility calculation formula generation process)
The process in which the moving body 10 generates the credibility calculation formula will be described. As shown in FIG. 6, the arithmetic unit 18 includes a credibility calculation formula generation unit 46 and a credibility calculation formula storage unit 48 in addition to the parts used in the above-described processing.

（Ｚ：目的変数、Ｘ_１〜Ｘ_３：説明変数、ｂ_０：定数、ｂ_１〜ｂ_３：回帰係数）に学習データを適用して、定数ｂ_０及び回帰係数ｂ_１〜ｂ_３の値を求める。学習データを適用する際は、説明変数Ｘ_１、Ｘ_２、Ｘ_３に学習データの標準偏差σ_ｘ、標準偏差σ_ｙ、最大確率ｗ_ｍａｘのそれぞれ用いる。これにより、次の数式で表される信憑度計算式；

が生成される（図７参照）。ｐの値は、第１計算部２８により計算される推定位置がＮＧとなる確率である。ｐはまた、計算された推定位置の真値からのずれ量の大きさを示す尺度でもあり、ずれ量が大きいほど大きくなる。本明細書ではｐを不信度という。不信度ｐが所定値よりも大きくなったことは、計算した推定位置の信憑度が所定値より低下したことを意味する。
説明変数Ｘ_１〜Ｘ_３に用いる変数（σ_ｘ、σ_ｙ、ｗ_ｍａｘ）は互いに独立しているため、回帰係数ｂ_１〜ｂ_３は、項目間の影響が排除された係数となる。従って、不信度を算出する際に、より正確に項目ごとの影響度を量ることができる。信憑度計算式生成処理はＣＰＵ２０で行われる。
信憑度計算式記憶部４８は、ＲＯＭ２４に設けられており、信憑度計算式生成部４６で生成された信憑度計算式を記憶する。 The credibility calculation formula generation unit 46 acquires the learning data stored in the learning data storage unit 38, performs logistic regression analysis on the learning data, and generates a credibility calculation formula. Logistic regression analysis is a method of multivariate analysis, and is a method of generating a statistical regression model that can predict the probability that the result will occur when the result is binary. Specifically, the logistic model expressed by the following formula;

Applying the training data to (Z: objective variable, X _{1 to} X ₃ : explanatory variable, b ₀ : constant, b _{1 to} b ₃ : regression coefficient), the values of the constant b ₀ and the regression coefficients b _{1 to} b ₃ Ask for. When applying the training data, the standard deviation σ _x , the standard deviation σ _y , and the maximum probability w _max of the training data are used for the explanatory variables X ₁ , X ₂ , and X ₃ , respectively. As a result, the credibility calculation formula expressed by the following formula;

Is generated (see FIG. 7). The value of p is the probability that the estimated position calculated by the first calculation unit 28 will be NG. p is also a measure of the amount of deviation of the calculated estimated position from the true value, and the larger the amount of deviation, the larger the amount. In this specification, p is referred to as distrust. When the distrust p becomes larger than the predetermined value, it means that the credibility of the calculated estimated position is lower than the predetermined value.
Since the variables (σ _x , σ _y , w _max ) used for the explanatory variables X _{1 to} X ₃ are independent of each other, the regression coefficients b _{1 to} b ₃ are coefficients excluding the influence between the items. Therefore, when calculating the distrust, the degree of influence for each item can be measured more accurately. The credibility calculation formula generation process is performed by the CPU 20.
The credibility calculation formula storage unit 48 is provided in the ROM 24, and stores the credibility calculation formula generated by the credibility calculation formula generation unit 46.

(Credibility calculation processing)
The process in which the moving body 10 calculates the distrust p is described. The arithmetic unit 18 includes a credibility calculation unit 50 in addition to the units used in the above-described processing (see FIG. 2).

When the moving body 10 autonomously travels in the autonomous driving region, the first calculation unit 28 calculates the estimated position using the autonomous driving environment map (second environment map). The three variables σ _x , σ _y , and w _max when the current estimated position is calculated by the first calculation unit 28 are output to the credibility calculation unit 50.

The credibility calculation unit 50 acquires the three variables σ _x , σ _y , and w _max obtained from the first calculation unit 28 from the credibility calculation formula storage unit 48, and obtains the credibility calculation formulas X ₁ , X ₂ , and X. Enter each in ₃ to find the distrust p in the current step. The distrust p is calculated step by step. This process is performed by the CPU 20. The distrust degree p calculated by the credibility calculation unit 50 is output to the map update determination unit 58 (described later). In the first step after the moving body 10 starts moving from the starting point, the distrust degree p, which is a criterion for determination, has not yet been calculated. The calculation of the distrust p is carried out from the second and subsequent steps.
The distrust degree p is also transmitted to the weighting coefficient calculation unit 52. This will be described later.

(Calculation processing of estimated position independent of map)
As shown in FIG. 2, in addition to the first calculation unit 28 that calculates the estimated position depending on the map, the calculation device 18 is a map-independent estimated position calculation unit (second) that calculates the estimated position that does not depend on the map. 29 (abbreviated as a calculation unit), an integrated estimation position calculation unit 54, and an integrated estimation position storage unit 56. The integrated estimated position storage unit 56 is provided in the RAM 22, and stores the integrated estimated position (strictly speaking, the coordinates on the environment map) estimated by the integrated estimated position calculation unit 54 for each step. The second calculation unit 29 adds the movement amount vector for one step of the moving body 10 to the integrated estimation position in the previous step among the integrated estimation positions stored in the integrated estimation position storage unit 56. The value calculated by is estimated as the position of the moving body 10 in the current step. The movement vector is available from the encoder 12. As is clear from the above description, the second calculation unit 29 calculates the estimated position without directly using the environment map. The estimated position calculated by the second calculation unit 29 may include an element depending on the environment map. This is because the integrated estimated position in the previous step contains an element that depends on the map.

In addition, the above-mentioned "integrated estimated position in the previous step" is, in other words, the newest integrated estimated position among the integrated estimated positions stored in the integrated estimated position storage unit 56. When the current step is the first step after the moving body 10 starts moving from the start point, the "integrated estimated position in the previous step" means the position of the start point.

(Integrated estimated position calculation processing)
The mobile body 10 performs the above-mentioned learning data generation process and credibility calculation formula generation process before autonomous driving in the second region. Then, when the moving body 10 autonomously travels, the weighting coefficient is calculated based on the credibility calculated from the credibility calculation formula, and the integrated estimated position is calculated using the weighting coefficient. Hereinafter, the integrated estimated position calculation process during autonomous driving of the moving body 10 will be described. As shown in FIG. 2, the arithmetic unit 18 includes a weight coefficient calculation formula storage unit 51, a weight coefficient calculation unit 52, and an integrated estimation position calculation unit 54, in addition to each unit used in the above-described processing.

（ｐ：不信度、α：重み係数）で表される（図８参照（後述））。
重み係数計算部５２は、重み係数計算式記憶部５１から取得した重み係数計算式に、信憑度計算部５０から取得した現在のステップにおける不信度ｐを入力して重み係数αを求める。この処理はＣＰＵ２０で行われる。重み係数計算部５２で算出された重み係数αは、統合推定位置計算部５４に出力される。 The weight coefficient calculation formula storage unit 51 is provided in the ROM 24 and stores a preset weight coefficient calculation formula. The weighting coefficient calculation formula is the following sigmoid function;

It is represented by (p: distrust, α: weighting coefficient) (see FIG. 8 (described later)).
The weighting coefficient calculation unit 52 inputs the distrust degree p in the current step acquired from the credibility calculation unit 50 into the weighting coefficient calculation formula acquired from the weighting coefficient calculation formula storage unit 51 to obtain the weighting coefficient α. This process is performed by the CPU 20. The weighting coefficient α calculated by the weighting coefficient calculation unit 52 is output to the integrated estimation position calculation unit 54.

（ｒ_ｆ：統合推定位置、ｒ_１：地図に依存して推定した位置、ｒ_２：地図に依存しないで推定した位置）に従って計算を行い、移動体１０の現在のステップにおける統合推定位置を算出する。重み係数αは「地図に依存する推定位置」の重み係数として使用され、１から重み係数αを減じた係数１−αは「地図に依存しない推定位置」の重み係数として使用される。この処理は、ＣＰＵ２０で行われる。統合推定位置計算部５４で算出された統合推定位置は、統合推定位置記憶部５６と観測点群生成部６０（後述）に出力される。統合推定位置記憶部５６に出力されて記憶される統合推定位置は、１つ後のステップにおける地図に依存しない推定位置の計算処理に用いられる。 The integrated estimated position calculation unit 54 uses the weight coefficient α acquired from the weight coefficient calculation unit 52 to acquire the estimated position (estimated position depending on the map) obtained from the first calculation unit 28 and the second calculation unit 29. The weighted average value of the estimated position (estimated position independent of the map) is calculated, and that value is used as the integrated estimated position. Specifically, the integrated estimation position calculation unit 54 uses the following mathematical formula;

(R _f: Integration estimated position, r _1: a position that is estimated depending on the map, r _2: the position estimated without depending on the map) performs calculation in accordance with the calculated integration estimated position at the current step of the moving body 10 To do. The weighting coefficient α is used as the weighting coefficient of the “map-dependent estimated position”, and the coefficient 1-α obtained by subtracting the weighting coefficient α from 1 is used as the weighting coefficient of the “map-independent estimated position”. This process is performed by the CPU 20. The integrated estimated position calculated by the integrated estimated position calculation unit 54 is output to the integrated estimated position storage unit 56 and the observation point cloud generation unit 60 (described later). The integrated estimated position output and stored in the integrated estimated position storage unit 56 is used for the calculation process of the estimated position independent of the map in the next step.

Here, the weighting coefficient calculation formula will be described with reference to the above function (Equation 3) and FIG. The weighting coefficient α calculated by the weighting coefficient calculation formula is a function of the distrust degree p. The weighting coefficient α, which is the weighting coefficient of the estimated position depending on the map, decreases as the distrust p increases (that is, as the estimation accuracy of the estimated position depending on the map decreases). Specifically, the value of the weighting coefficient α is close to 1 when the distrust p satisfies 0 <p <0.1, 0.5 when p = 0.2, and 0. When 3 <p is satisfied, it is close to 0. Therefore, the integrated estimated position is almost the same as the "map-dependent estimated position" when the distrust p is less than 10%, and is called the "map-dependent estimated position" when the distrust p is 20%. It is an intermediate point of the "estimated position that does not depend on the map", and when the distrust rate exceeds 30%, it almost coincides with the "estimated position that does not depend on the map". The value of the constant used in the weighting coefficient calculation formula is not limited to the above values (50, 0.2). Moreover, the weighting coefficient calculation formula is not limited to the sigmoid function. Various functions can be used depending on how the value of the weighting coefficient α is associated with the value of the distrust p.

(Map update judgment process)
Next, the map update determination process will be described. When the moving body 10 travels in the vicinity of an obstacle that is not recorded in the autonomous driving environment map (second environment map) in the autonomous driving area, the observation result acquired by the LRF 14 and the calculation result calculated from the environment map. Since the degree of matching with is lowered, the estimation accuracy of the "estimated position depending on the map" is lowered. Therefore, the distrust p is high. That is, even if the true value is included in the particle set, the particle corresponding to the true value of the moving body 10 cannot be regarded as the conforming particle, and another particle is regarded as the conforming particle. In the moving body 10 of the embodiment, the first calculation unit 28 calculates the estimated position depending on the map according to the method of MCL. Therefore, when the moving body 10 travels near an obstacle that is not recorded on the second environmental map, the estimated position depending on the map calculated by the first calculation unit 28 is separated from the true value. Eventually, the particle set will not contain the true value. Once the particle set no longer contains the true value, even if the moving body 10 subsequently travels in the autonomous traveling region (second region) in the region where the correct environmental map is stored (that is,). Even if it becomes possible to obtain useful observation results), it may not be possible to generate a particle set in the range containing the true value by the MCL method. That is, it may not be possible for the first calculation unit 28 to accurately calculate the estimated position depending on the map. In order to avoid such a situation, the map update determination unit 58 outputs an environmental map update command to the observation point cloud generation unit 60 when the distrust p is increasing and the distrust p exceeds 20%. .. That is, when the distrust degree p satisfies 0 <p ≦ 0.2 (when the credibility of the estimated position depending on the map is high), the map update determination unit 58 does not output the update command of the environment map and is 0. When .2 <p is satisfied (when the credibility of the estimated position depending on the map is low), the update command of the environment map is output to the observation point group generation unit 60. This process is performed by the CPU 20. In this embodiment, the value at which the weight of the first calculation unit 28 and the weight of the second calculation unit 29 are 50:50 (that is, p = 0.2) is set as the threshold value of the update command of the environment map. The distrust p that serves as a criterion for determining whether or not to update the environment map is not limited to the above. For example, it may be less than 20% or more than 20%.

(Environmental map update process)
Next, an example of the process in which the arithmetic unit 18 updates the environment map will be described with reference to FIGS. 9 and 10. First, the map point cloud generation unit 62 converts the environment map (see A in FIG. 10) stored in the moving body 10 into a map point cloud (see B in FIG. 10) represented by xy coordinate values. (S10). Next, the observation point cloud generation unit 60 converts the current observation result by the LRF 14 into an observation point cloud (see C in FIG. 10) represented by the xy coordinate value (S12). The integrated estimation position in the current step is input to the observation point cloud generation unit 60 from the integrated estimation position calculation unit 54. The observation point cloud generation unit 60 generates an observation point cloud by converting the observation result by the LRF 14 into the environment map coordinate system (xy coordinate system) based on this integrated estimated position. The map point cloud and the observation point cloud converted in steps S12 and S14 are output to the point cloud matching unit 64, respectively. The arithmetic unit 18 may be configured so that the processes of steps S12 and S14 are executed first.

Next, the point cloud matching unit 64 aligns the map point cloud and the observation point group (S16). An ICP (Iterative Closet Point) algorithm can be used to match the point cloud. ICP is a technique for matching by minimizing the sum of squares of the distances of two different point clouds. Since ICP is a known technique, detailed description thereof will be omitted. The point cloud matched by ICP (see D in FIG. 10) is output to the environment map update unit 66.

When the observation point cloud and the map point cloud are matched, a region representing the change in the environment appears. In reality, there may be obstacles that are recorded but not observed on the environmental map, and obstacles that are not observed on the environmental map but are observed. The environment map update unit 66 generates the latest environment map by reflecting each change in the environment map stored in the moving body 10. Specifically, first, the environment map update unit 66 generates a free grid (see E in FIG. 10) in order to reflect the object removed from the current environment map stored in the moving body 10 on the grid map. (S16). A free grid is a grid on a grid map that represents the absence of obstacles, between the current position (ie, the integrated estimated position of the current step) and the observation point cloud observed by LRF14. Space is made into a free grid. That is, the ICP grids the observation point groups that match the map point group, and creates a free grid as an obstacle (blank area) from the position (position) of the moving body 10 to the observation point group.

Next, the environment map update unit 66 generates an occupied grid (see F in FIG. 10) in order to reflect the obstacle added to the current environment map stored in the moving body 10 on the grid map (S18). .. The occupied grid is a grid on a grid map showing the existence of an object, and a point cloud of changes that does not exist in the map point cloud but exists in the observation point cloud is converted into an occupied grid. That is, by matching by ICP, a point cloud (difference point cloud) that does not exist in the map point cloud but exists in the observation point cloud is extracted, and the extracted point cloud is gridded to generate an occupied grid.

Next, the environment map update unit 66 generates the latest environment map (see G in FIG. 10) by superimposing the generated free grid and occupied grid on the environment map stored in the moving body 10 (S20). .. By executing the above processing, the latest environment map is generated. The environment map update unit 66 ends the process by updating the environment map currently stored in the moving body 10 to the latest environment map (S22).

The updated environmental map (latest environmental map) is used again when the first calculation unit 28 estimates the estimated position depending on the map. At this time, the observation result by the LRF 14 is accurately matched with the latest environmental map, and the distrust of the first calculation unit 28 is reduced (the credibility is improved). Therefore, the map update command by the map update determination unit 58 is not output, and the integrated estimation position calculation unit 54 emphasizes the estimated position depending on the map (that is, the distrust p is less than 20% in this embodiment). The integrated estimated position is estimated. The mode of the environment map update process is not limited to the above ICP. Various known methods can be used as long as the process of updating the environment map is started when the credibility drops below a predetermined value.

The action and effect of the moving body 10 of the example will be described. In the moving body 10, when the credibility of the estimated position depending on the map is lower than the predetermined value (when the distrust p is higher than the predetermined value), the distance to the obstacle and the azimuth angle are observed. Generate the latest environmental map. That is, when the credibility of the estimated position of the first calculation unit 28 decreases, the process of updating the environment map is performed. Therefore, it is possible to determine whether or not the environment map can be updated by the moving body 10 itself without requiring an operator's operation.

Moreover, since the point cloud converted to the coordinate system is used instead of the grid map when the observation result and the environment map are matched, the error due to the matching can be reduced. Further, by estimating the integrated estimated position using the mixed result of two types of estimated positions (estimated position depending on the map and estimated position not dependent on the map), the observation point group can be generated with high accuracy. Therefore, the dissociation between the observation point cloud and the map point cloud becomes small, and the accuracy of point cloud matching by ICP can be improved.

Further, the weighting coefficient α used when the integrated estimation position calculation unit 54 estimates the integrated estimation position of the moving body 10 is a function of the distrust p. The credibility calculation formula for calculating the distrust p is a formula obtained by analyzing the learning data acquired by the moving body 10 learning and running by using a method of logistic regression analysis which is a method of machine learning. Is. Therefore, the weighting coefficient α is not affected by factors other than the credibility calculation formula. Therefore, for example, even if the moving area is changed or the administrator of the moving body 10 is changed, the standard for determining the value of the weighting coefficient α does not change. As a result, the estimation accuracy of the integrated estimated position of the moving body 10 can be stabilized.

The estimated position that does not depend on the map is obtained by adding the movement vector for one step calculated based on the rotation angles of the left and right wheels acquired from the encoder 12 to the integrated estimated position in the previous step. Desired. The information detected by the encoder 12 may include an error due to slippage of each wheel or the like, but when the detection distance by the encoder 12 is relatively short, the above error can be almost ignored. Since the movement vector for one step is extremely short, the accuracy of the movement vector for one step is high. In addition, the estimated position that does not depend on the map is estimated without using the environmental map. Therefore, even when the moving body 10 autonomously travels near an obstacle that is not recorded on the environmental map, the estimation accuracy of the estimated position that does not depend on the map does not decrease due to this. .. If the estimation accuracy of the integrated estimated position one step before is high, the estimated position that does not depend on the map in the current step can maintain high estimation accuracy.

In the moving body 10 of the present embodiment, as the distrust p increases, the map-dependent weighting coefficient α of the estimated position decreases, and the map-independent estimated position weighting coefficient 1-α increases. According to this configuration, when the distrust p is high (that is, when the estimation accuracy of the estimated position depending on the map is low), the ratio of the estimated position depending on the map to the integrated estimated position is reduced, and the estimated position is not dependent on the map. The percentage of estimated positions can be increased. As described above, the estimation accuracy of the estimated position that does not depend on the map is high. Therefore, according to this configuration, even if the estimation accuracy of the estimated position depending on the map is lowered, it is possible to suppress the deterioration of the estimation accuracy of the integrated estimated position. Further, since the weighting coefficient calculation formula is a continuous and smooth function, the weighting coefficient α fluctuates smoothly according to the transition of the distrust p. Therefore, the integrated estimated position is a value that can be continuously changed between the estimated position that depends on the map and the estimated position that does not depend on the map. Therefore, the estimation accuracy of the integrated estimation position is further improved.

In this embodiment, as described above, the estimated result of the second calculation unit 29, which calculates the estimated position independent of the map, is advantageous for the integrated estimated position when the map update determination unit outputs the update command of the environmental map. It has become. Therefore, the latest environment map can be generated with high accuracy.

Further, in this embodiment, the information that is the basis of the learning data (that is, variables such as the position estimation result and the standard deviation σ _x ) is acquired by actually running the moving body 10. Therefore, the learning data directly reflects the calculation performance of the estimated position depending on the map of the moving body 10. In this way, by generating the learning data based on the information obtained by actually running the vehicle, it is possible to generate a more accurate credibility calculation formula.

Although the examples of the techniques disclosed in the present specification have been described in detail above, these are merely examples, and the moving bodies disclosed in the present specification include various modifications and modifications of the above-described examples. Is done.

For example, in the above-described embodiment, the position is estimated by the first calculation unit 28 and the second calculation unit 29, but the position may be estimated by using an estimation method other than these. For example, the position can be estimated by using a camera or GPS.

Further, in the above-described embodiment, the average value of the estimated position depending on the map and the estimated position not dependent on the map is set as the integrated estimated position, and the process of updating the environment map based on the integrated estimated position is executed, but this is limited to this. Absent. For example, when the credibility of the first calculation unit 28 that calculates the estimated position depending on the map is lower than the predetermined value, the estimated position depending on the map at the last time when the credibility is equal to or higher than the predetermined value (that is, the estimated position depending on the map). The estimated position at the present time is the position obtained by adding the movement vector for one step to the estimated position (estimated position depending on the map immediately before the credibility becomes less than the predetermined value), and based on the estimated position at the present time and the observation result. , You may perform the process of generating the latest environment map. Further, when the credibility of the estimated position by the first calculation unit 28 is lower than the predetermined value, the estimated position depending on the map at the time when the credibility is equal to or higher than the predetermined value and the observation result observed at that time are obtained. Based on this, the latest environmental map may be generated. Even with such a configuration, it is possible to obtain the same effect as that of the moving body 10 of the embodiment. In the case of the latter method, it is not necessary to calculate the estimated position that does not depend on the map, and only the estimated position that depends on the map can be calculated. In this case, the estimated position depending on the map may be the integrated estimated position.

Further, in the first calculation unit 28, an extended Kalman filter may be used instead of the MCL to calculate the estimated position depending on the map. In this case, the credibility calculation formula is generated by using the values obtained from the estimated error covariance matrix (for example, the variance value, the value of each element of the error covariance matrix, the major axis and the minor axis of the error ellipse, etc.) as explanatory variables. can do. Further, by inputting these variables obtained during autonomous driving into the credibility calculation formula, it is possible to accurately determine the estimated position estimation result depending on the map.

Further, in the above embodiment, the movement information is acquired by using the encoder 12 provided in the motor, but the method of acquiring the movement information is not limited to this. For example, movement information may be acquired using visual odometry using a camera or laser odometry using a laser sensor.

Further, the moving body 10 does not have to have the learning data generation unit 36, and the learning data may be generated outside the moving body 10. Further, the moving body 10 does not have to have the learning data storage unit 38. That is, the learning data may be deleted when the credibility calculation formula is generated based on the learning data. Further, the mobile body 10 does not have to have the credibility calculation formula generation unit 46, and the credibility calculation formula may be generated outside the mobile body 10.

Further, in the credibility calculation formula generation unit 46, a support vector machine may be used instead of the multivariate analysis. The support vector machine is a known method in modeling pattern recognition, and is a method of constructing an identification function that outputs a class to which a class belongs as a binary value when a feature vector whose class attribution is unknown is input. The discriminant function consists of a set of training samples with known class attribution. Therefore, the discriminant function (credibility calculation formula) can be constructed by using the training data of the examples as a sample set for training and learning the training data with a support vector machine. Then, by inputting the three variables σ _x , σ _y , and w _max acquired from the first calculation unit 28 into the identification function, the estimated position estimation result depending on the map can be determined.

Further, in the credibility calculation formula generation unit 46, a neural network may be used instead of the multivariate analysis. A neural network is a known method in modeling pattern recognition, and is a method of constructing an identification function that outputs a class to which a class belongs as a binary value when a variable whose class attribution is unknown is input. In this method, the discriminant function can be constructed in two ways, supervised learning and unsupervised learning. In supervised learning, a learning example and a target output for the learning example are given, and the weight of the function in the initial state is adjusted so that the output obtained by inputting the learning example into the function in the initial state matches the target output. By doing so, the identification function is constructed. Therefore, by learning the training data with a neural network using the three variables σ _x , σ _y , and w _max of the training data of the example as the learning example and the classification result (OK / NG) at that time as the target output. , Discrimination function (credibility calculation formula) can be constructed. Then, by inputting the three variables σ _x , σ _y , and w _max acquired from the first calculation unit 28 into the identification function, the estimated position estimation result depending on the map can be determined.

Further, when analyzing the training data, discriminant analysis may be used instead of logistic regression analysis.

Further, the learning run may be performed by simulation instead of actually causing the moving body to run in a specific area.

Further, the moving body 10 does not have to be a wheel-driven type, and may be, for example, a small mobility type that moves at a low speed, a leg type (bipedal walking, etc.), or a flying body. Further, the moving body 10 may have a plurality of LRF14s.

Further, the true value calculation unit 32 calculates the true value of the moving body 10 by using the triangulation sensor 16 and the reflector 44, but the method of calculating the true value is not limited to this. For example, the true value may be calculated using motion capture or a magnetic rail technique.

Further, in this embodiment, a two-dimensional observation result is used for position estimation or the like, but a three-dimensional observation result obtained by a three-dimensional distance sensor or the like may be used.

Further, the environment map stored in the moving body 10 may be stored as a point cloud represented by coordinate values instead of a grid map. According to this configuration, the latest environmental map can be generated even if the map point cloud generation unit 62 in FIG. 2 is not provided.

In the technique disclosed in the present specification, the map-dependent estimation position calculation means calculates the probability that the autonomous moving body exists at the assumed position for each assumed position (particle), and estimates the position from the probability distribution. .. The credibility calculation means calculates the credibility from the probability distribution. The credibility referred to here is the probability that the estimated position remains within the permissible range from the true value.

The autonomous moving body disclosed in the present specification further includes a map-independent position estimating means for estimating the position of the autonomous moving body from the moving distance and the moving direction. In this case, when the credibility is lowered and the environment map is updated, the last calculated estimated position in the period when the credibility is equal to or higher than the predetermined value (this estimated position may be an estimated position based on the map). , The estimated position that does not depend on the map and the estimated position that does not depend on the map may be integrated.), And the latest position is obtained by adding the subsequent movement amount and movement direction, and the latest position and the observation device are the last. The environmental map is updated based on the information on the distance and direction to the obstacle observed in. Instead, when the credibility drops below the specified value, the last calculated estimated position within the period when the credibility is above the specified value, and the distance and orientation information to the obstacle observed by the observation device at that time. Therefore, the environment map may be updated.

In the autonomous mobile body disclosed in the present specification, since the environment map is updated when the update is necessary, the credibility of the estimated position depending on the map is high. Only the calculation unit of the estimated position that depends on the map is sufficient, and the calculation unit of the estimated position that does not depend on the map can be omitted.
The autonomous moving body of this embodiment has both a map-dependent estimated position calculation unit and a map-independent estimated position calculation unit, and integrates both to calculate the integrated estimated position. At that time, it is determined which estimated position is to be emphasized based on the credibility of the estimated position calculation unit of the map-dependent formula. If the credibility of the map-dependent estimation position calculation unit is high, the one that emphasizes the map-dependent estimation position is set as the integrated estimation position, and if the credibility of the map-dependent estimation position calculation unit is low, the map-independent estimation is performed. The position-oriented position is the integrated estimated position. The credibility of the integrated estimated position is high, and the credibility of the position of the autonomous mobile body used when updating the environmental map is high.

The environmental map update device can be made independent, and the environmental map update device itself can be manufactured, sold, and used.

The relationship between the components of the embodiment and the components of the claims will be described. The environmental map storage unit 26 of the embodiment is an example of the environmental map storage means of the claim. The LRF14 of the embodiment is an example of the observation device of the claim. The first calculation unit 28 of the embodiment is an example of the estimation position calculation means of the map-dependent formula of the claim. The second calculation unit 29 of the embodiment is an example of the estimation position calculation means of the map-independent equation of the claim. The credibility calculation unit 50 of the embodiment is an example of the credibility calculation means of the claim. The observation point cloud generation unit 60, the map point cloud generation unit 62, the point cloud combination unit 64, and the environmental map update unit 66 of the embodiment are examples of the environmental map update means according to the claim.

Although specific examples of the present invention have been described in detail above, these are merely examples and do not limit the scope of claims. The techniques described in the claims include various modifications and modifications of the specific examples illustrated above. In addition, the technical elements described in the present specification or the drawings exhibit technical usefulness alone or in various combinations, and are not limited to the combinations described in the claims at the time of filing. In addition, the techniques illustrated in the present specification or drawings achieve a plurality of objectives at the same time, and achieving one of the objectives itself has technical usefulness.

10: Mobile 12: Encoder 14: Laser range finder (LRF)
26: Environmental map storage unit 28: First calculation unit (map-dependent estimation position calculation unit)
29: Second calculation unit (map-independent estimation position calculation unit)
46: Confidence calculation formula generation unit 48: Confidence calculation formula storage unit 50: Confidence calculation unit 51: Weight coefficient calculation formula storage unit 52: Weight coefficient calculation unit 54: Integrated estimation position calculation unit 56: Integrated estimation position storage unit 58: Map update judgment unit 60: Observation point group generation unit 62: Map point group generation unit 64: Point group adjustment unit 66: Environmental map update unit

Claims (9)

It is an autonomous mobile body
An environmental map storage means that stores an environmental map showing the positions of obstacles existing in the area where the autonomous moving body moves, and
An observation device that observes the distance and direction from the autonomous moving body to the obstacle,
Assumed position where the autonomous moving body is located on the environmental map, information on the distance and direction to the obstacle calculated from the environmental map, and information on the distance and direction to the obstacle observed by the observation device. From the map-dependent estimation position calculation means for estimating the position of the autonomous moving body,
The credibility calculation means for calculating the credibility of the estimated position by the map-dependent position estimation means, and the credibility calculation means.
When the credibility drops below a predetermined value, the environment map update that updates the environment map stored by the environment map storage means based on the information of the distance and the direction to the obstacle observed by the observation device. means,
An autonomous mobile that is equipped with. Wherein the map-dependent estimated position calculating means, the autonomous moving body for each of the presumed position calculates the probability that exists in its presumed position, in Claim 1, characterized in that to estimate the position from the distribution of the probability The described autonomous mobile body. The autonomous mobile body according to claim 2, wherein the credibility calculating means calculates the credibility from the distribution of the probabilities. The autonomous mobile body according to claim 3, wherein the credibility calculation means calculates the credibility at which the estimated position remains within an allowable range from the true value from the distribution of the probabilities. The autonomous moving body further includes a map-independent position estimating means for estimating the position of the autonomous moving body from the moving distance and the moving direction.
When the credibility is lower than the predetermined value, the latest position obtained by adding the subsequent movement amount and the movement direction to the last estimated position where the credibility is equal to or higher than the predetermined value is obtained, and the latest position and the observation are obtained. The autonomous moving body according to any one of claims 1 to 4, wherein the environmental map is updated from the information of the distance and the direction to the obstacle observed by the apparatus. When the credibility is lower than the predetermined value, the distance and direction between the last estimated position where the credibility is equal to or higher than the predetermined value and the obstacle observed by the observation device at the last estimated position. The autonomous moving body according to any one of claims 1 to 4, wherein the environmental map is updated from the information. The autonomous mobile body
A map-independent estimated position calculation means for estimating the position of the autonomous moving body from the moving distance and the moving direction, and
It is equipped with a device for calculating the integrated estimated position by weight averaging the estimated position by the map-dependent estimated position calculating means and the estimated position by the map-independent estimated position calculating means.
The autonomous mobile body according to any one of claims 1 to 6, wherein the weight used for the weight averaging is determined based on the credibility. According to any one of claims 1 to 7, the estimation position calculation means of the map-dependent expression estimates the position by using either the Monte Carlo position identification method (MCL) or the extended Kalman filter method. The described autonomous mobile body. It is an environmental map updater for autonomous mobiles that is equipped with an observation device that observes the distance and direction to the obstacle and stores an environmental map that shows the position of the obstacle existing in the moving area.
Assumed position where the autonomous moving body is located on the environmental map, information on the distance and direction to the obstacle calculated from the environmental map, and information on the distance and direction to the obstacle observed by the observation device. From the map-dependent estimation position calculation means for estimating the position of the autonomous moving body,
The credibility calculation means for calculating the credibility of the estimated position by the map-dependent estimation position calculation means, and the credibility calculation means.
When said authenticity degree is lower than a predetermined value, the observation device is based on the distance and azimuth information to the obstacle observed before Symbol environment map update means for updating the environmental map,
Environmental map updater equipped with.

Images