JP5120024B2 - Autonomous mobile robot and obstacle identification method thereof - Google Patents

Description

  The present invention relates to an autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle, and an obstacle identification method thereof.

  Autonomous mobile robots need advanced situation recognition in order to grasp the current position, avoid obstacles, generate a route to the destination, and the like. In particular, an autonomous mobile robot operating in a real environment needs to identify what kind of object is a surrounding obstacle in a complicated situation.

  Conventionally, in autonomous mobile robots, methods such as matching by measuring the shape of a target object using a stereo camera or other three-dimensional coordinate measurement technique are used (for example, see Patent Documents 1 to 3 below). However, when a camera image is used, the field of view is not always good in a complicated environment, and there is a problem that it is expensive when an advanced sensor such as a laser range finder is used.

  Therefore, in an autonomous mobile robot that is actually operated in a complex environment, it is important that the autonomous mobile robot can acquire necessary information using a simple sensor.

JP 2004-326264 A JP 2004-280451 A JP 2006-079157 A

  The present invention has been made in view of the above-described problems. The object of the present invention is to use only a distance sensor, and any obstacle that reacts to the distance sensor can be a person, a wall, or another obstacle. An object of the present invention is to provide an autonomous mobile robot capable of estimating whether there is an obstacle and a method for identifying the obstacle.

In order to achieve the above object, according to one aspect of the present invention, an autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle, the distance from the obstacle is set. On the probability model , based on the distance measurement means for measuring, the distance history evaluation means for evaluating the history of the distance information measured by the distance measurement means, and the evaluation result by the robot behavior and the distance history evaluation means Obstacle inference means for inferring an obstacle by performing probability inference, and robot behavior that enables the distance measurement means to acquire high-value distance information when the obstacle inference means performs efficient probability inference. An autonomous mobile robot comprising observation behavior determining means for determining is provided.

  According to another aspect of the present invention, there are provided an obstacle identification method executed by the above-described autonomous mobile robot, and a program for causing a computer system to function as the above-described autonomous mobile robot.

  According to the disclosed autonomous mobile robot and its obstacle identification method, based on only time-series distance information from the distance measurement means, it is inferred what kind of obstacle is the obstacle that reacts to the distance measurement means. It becomes possible.

  Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings. FIG. 1 is a plan view showing an overview of an autonomous mobile robot 100 according to an embodiment of the present invention. The robot 100 includes a left front distance sensor 102a, a front distance sensor 102b, and a right front distance sensor 102c (hereinafter collectively referred to as reference numeral 102). These distance sensors are, for example, laser radars. Each of the left front distance sensor 102a and the right front distance sensor 102c is directed in a direction shifted leftward or rightward by 28 ° with respect to the direction of the front distance sensor 102b.

  The robot 100 also includes a moving mechanism and a warning mechanism (not shown) including moving wheels and a drive motor. Then, the robot 100 takes five actions: “still”, “forward”, “right turn”, “left turn”, and “warning”.

  Then, for example, the robot 100 performs the following inference. That is, when the value of the distance sensor 102 monotonously decreases during the turning action, the robot 100 changes from the direction of the distance sensor before turning (the direction A) to the direction of the distance sensor after turning (the direction B). It is inferred that there is a high possibility that there is a wall in the A direction because the distance is far away. In addition, when the time series information of the distance sensor 102 greatly fluctuates during the stationary action, the robot 100 infers that there is a high possibility that there is a moving obstacle such as a person in that direction.

  In this way, in inferring the situation using only the time series information from the distance sensor 102, the robot 100 specifically adopts the following method. That is, the robot 100 creates a probability model of the situation and the observation predicted for each robot action. Then, the robot 100 infers the probability distribution of the situation at each time based on the behavior at each time and the obtained observation.

  The robot 100 uses, for example, information on the presence / absence of an obstacle detected by each sensor, trend of sensor input (distance) history at the time of performing each action (stationary, increase / decrease, other changes) from the measurement information obtained by the distance sensor 102. Etc.) is used as actual observation information. Also, the robot 100 selects an action with a high possibility of obtaining information necessary for inference and performs active observation in order to infer efficiently.

  The robot 100 uses a dynamic Bayesian network in order to perform inference by integrating observation information of each step as a method of performing probability inference on a probability model. The details will be described below.

  FIG. 2 is a diagram illustrating a functional configuration of the autonomous mobile robot 100. The robot 100 is a robot that moves autonomously while recognizing the obstacle 220 as an observation target. In addition to the distance sensor 102 and the movement mechanism 202, the robot 100 includes a distance history evaluation unit 204, an obstacle reasoning unit 206, and the like. An observation behavior determination unit 208 and an autonomous movement control unit 210 are functionally provided.

  The robot 100 is a computer system having a distance sensor 102 and a moving mechanism 202 as an input device and a control target, respectively, and includes a processor, a memory, and the like (not shown). Then, the functions of the distance history evaluation unit 204, the obstacle reasoning unit 206, the observation behavior determination unit 208, and the autonomous movement control unit 210 are realized by the processor operating according to the program loaded in the memory. These functions will be described below, but before that, the Bayesian network will be described.

  FIG. 3 is a diagram for explaining a Bayesian network, in which (a) shows an acyclic directed graph, and (b), (c), and (d) are attached to node A, node B, and node C, respectively. A probability table is shown. A Bayesian network is a network-like probability model defined by three components: a random variable, a conditional dependency relationship between the random variables, and a conditional probability in the conditional dependency relationship.

  A random variable is represented by a node. Conditional dependencies between random variables are represented by directed links (arrows) spanning between nodes, the node at the end of the link is called the child node, and the node at the link origin is called the parent node . The conditional probability in the conditional dependency relationship is provided with the node, and a conditional probability table (CPT: Conditional Table) that lists the probabilities that the node takes a certain value when the parent node of the node takes a certain value. Probability Table) A node having no parent node is accompanied by a probability table listing the probabilities (prior probabilities) for the node to take a certain value.

For example, FIG. 3A is an acyclic directed graph showing a Bayesian network in which the random variable C depends on the random variable A and the random variable B. FIG. 3B shows that the random variable A can take values a ₁ , a ₂ and a ₃ , and the prior probability distributions P (A) taking such values are 0.4, 0.5 and 0, respectively. .1 is a probability table showing that 1. Similarly, FIG. 3C is a probability table showing the prior probability distribution P (B) for the random variable B. FIG. 3D is a probability table showing a conditional probability distribution P (C | A, B) for a random variable C that depends on the random variable A and the random variable B. For example, A = a ₁ , and , When B = b ₁ , the probability that C = c ₁ is 0.7, the probability that C = c ₂ is 0.2, and the probability that C = c ₃ is 0.1. Yes.

  By using such a Bayesian network, it is possible to calculate a probability distribution for unobserved random variables when the values of some random variables are observed. Such a calculation is called probabilistic inference. In this way, if the probability distribution of the random variable desired to be known from the observed value can be calculated, the expected value of the subsequent random variable, the variable value having the maximum probability, the entropy, and the like can be obtained.

  Next, processing by the distance history evaluation unit 204 will be described. The distance history evaluation unit 204 evaluates the history of distance information measured by the distance sensor 102 as follows (1) to (8). However, in the following evaluation, T_sense is a threshold value that can be judged to have a significant response to the observation value of the distance sensor, T_static is static, T_flat is flat, and T_vary is a significant change amount. It is a threshold value and T_stataic <T_flat <T_vary.

(1) Get 3 inputs (d_1, d_2, d_3 in time series order) (2) If the input is an error even once, error (3) min (d_1, d_2, d_3)> T_sence No response (4) If max (d_1, d_2, d_3) −min (d_1, d_2, d_3) <T_static and max (d_1, d_2, d_3) <T_sence, then static (5) max ( d_1, d_2, d_3) −min (d_1, d_2, d_3) <T_flat and max (d_1, d_2, d_3) <T_sence, flat (6) max (d_1, d_2, d_3) −min (d_1 , d_2, d_3) <T_vary and max (d_1, d_2, d_3) <T_sence and d_1 <d_2 <d_3, monotonically increasing (7) max (d_1, d_2, d_3) −min (d_1, d_2, d_3) <T_vary, and max (d_1, d_2, d_3) <T_sence and d_1>d_2> d_3, monotonically decreasing (8) If none of the above apply, else

  Next, processing by the obstacle reasoning unit 206 will be described. FIG. 4 is a diagram simply showing a configuration of a Bayesian network as a probability model used by the obstacle reasoning unit 206 for probability inference. In FIG. 4, characters surrounded by solid lines such as “obstacle identification” and “behavior” indicate nodes, that is, random variables. An arrow indicates a dependency relationship between nodes.

  To explain each node, the node “obstacle identification” is a node that is an object of inference, and represents the situation of the obstacle in front of the robot, and the values of random variables are “wall”, “person”, “space”. ”Or“ Other obstacles ”.

  The node “behavior” represents an action for observation, and takes one of the values of “stationary”, “forward”, “right turn”, “left turn”, or “warning” as the value of the random variable.

  The node “posture” represents the direction of the robot, and takes one of the values “left”, “front”, and “right” as the value of the random variable.

  Each of the nodes “Left Forward Situation”, “Front Situation” and “Right Forward Situation” indicates whether there is an obstacle at that time in each direction and whether the obstacle is dynamic or static. One of the values “static obstacle”, “dynamic obstacle”, and “no obstacle” is taken.

  Each of the nodes “left continuity”, “left → front continuity”, “front → right continuity” and “right direction continuity” represents how the distance changes from the front to the left and right, The value of the random variable is any one of “constant”, “monotonically increasing”, “monotonically decreasing”, or “discontinuous”.

  Each of the nodes “presence / absence of left front sensor response”, “presence / absence of front sensor response” and “presence / absence of right front sensor reaction” indicates whether or not there is a sensor response within a certain distance. It takes one of the values “None” or “Error”.

  Each of the nodes “left front sensor history evaluation value”, “front sensor history evaluation value”, and “right front sensor history evaluation value” represents a value obtained by evaluating the sensor history at the time of executing the action, and the value of the random variable is “constant” ”,“ Monotonic increase ”,“ monotonic decrease ”,“ other ”or“ error ”.

  In this embodiment, a probability model is constructed by defining a causal relationship between nodes as shown in FIG. 4 and determining a conditional probability table. The obstacle reasoning unit 206 executes probability reasoning using this probability model. In the portion surrounded by the dotted line, the contents of the causality line and the probability table change depending on the selected action. In this model, the information that can be observed differs depending on the selected action, so that the action selection for observation becomes important. In this example, the inferred node is one of “obstacle identification”, but there may be a plurality of nodes. However, when deciding the action, it is necessary to decide which node's situation you want to know.

  Next, the behavior determination algorithm in the observation behavior determination unit 208 will be described. The selectable action is a∈A, the observation is o∈O, the situation is s∈S, the probability that the situation is s by the inference result one step before is p (s), and the probability model is p (o | s, a).

選択行動：

The observation behavior determination unit 208 performs the following calculation so that the distance sensor 102 can acquire distance information having high value for the obstacle reasoning unit 206 to perform efficient probability reasoning. Select an action. The selected action is transmitted to the autonomous movement control unit 210, and the autonomous movement control unit 210 controls driving of the movement mechanism 202 based on the selected action.
Expected average mutual information for each action:

Choice behavior:

  The probability model p (o | s, a) can be obtained, for example, by calculating an inference of a Bayesian network by a stochastic simulation method. Here, since the node to be inferred and the node to which the observation is input are fixed, the calculation can be performed offline in advance.

  Finally, the overall operation will be described. FIG. 5 is a flowchart showing the operation of the robot 100. First, when the operation starts, the obstacle reasoning unit 206 initializes the inference result p (s) of the situation one step before (block 502). This is because there is no inference result one step before immediately after the start.

  Next, the obstacle reasoning unit 206 calculates an expected value ExI (a) of the average mutual information amount for each action (block 504).

  Next, the observation behavior determination unit 208 determines the behavior a at the time t as described above based on the calculated ExI (a) (block 506).

  Next, the autonomous movement control unit 210 executes the action a by the robot, and the distance sensor 102 acquires the observation o (block 508).

Next, the obstacle reasoning unit 206 calculates the probability distribution p (s | o, a) of the inference result for each observation value when the behavior and the observation value are confirmed (block 510). p (s | o, a) can be calculated as follows.

  Next, the obstacle reasoning unit 206 determines the end of inference (block 512). When the termination condition is not satisfied, the obstacle reasoning unit 206 returns the inference result p (s | o, a) to the block 504 as the inference result of the previous step and proceeds to the next step (block 514). When the end condition is satisfied, the obstacle reasoning unit 206 ends the inference. The termination condition is determined when the maximum value of p (s | o, a) becomes sufficiently large and the average information amount of p (s | o, a) becomes sufficiently small.

  Regarding the above embodiment, the following supplementary notes are disclosed.

(Appendix 1) An autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle,
A distance measuring means for measuring the distance to the obstacle;
Distance history evaluation means for evaluating the history of distance information measured by the distance measurement means;
An obstacle inference means for inferring an obstacle based on the behavior of the robot and the evaluation result by the distance history evaluation means;
An autonomous mobile robot comprising:

  (Supplementary note 2) The autonomous mobile robot according to supplementary note 1, wherein the obstacle reasoning means infers an obstacle by performing probability reasoning on a probability model.

  (Additional remark 3) The observation additional action means which determines the action of the robot which the distance measurement means can acquire the distance information with high value in order for the obstacle reasoning means to perform efficient probability reasoning is further provided. 2. The autonomous mobile robot according to 2.

(Supplementary Note 4) An obstacle identification method in an autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle,
A distance measuring step in which the distance measuring means measures the distance to the obstacle; and
A distance history evaluation means for evaluating a history of distance information measured by the distance measurement step;
An obstacle inference means for inferring an obstacle based on the behavior of the robot and the evaluation result of the distance history evaluation step;
An obstacle identifying method for an autonomous mobile robot comprising:

  (Supplementary note 5) The obstacle identification method for an autonomous mobile robot according to supplementary note 4, wherein the obstacle inference step infers an obstacle by performing probability inference on a probability model.

  (Additional remark 6) The observation action determination means determines the action of the robot in which the distance measurement step can acquire high-value distance information when the obstacle inference step performs efficient probability reasoning. The obstacle identifying method for an autonomous mobile robot according to appendix 5, further comprising:

(Appendix 7) For an autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle, a computer system is
A distance measuring means for measuring the distance to the obstacle;
Distance history evaluation means for evaluating the history of distance information measured by the distance measurement means;
An obstacle inference means for inferring an obstacle based on the behavior of the robot and the evaluation result by the distance history evaluation means;
Program to function as.

  (Supplementary note 8) The program according to supplementary note 7, wherein the obstacle reasoning means infers an obstacle by performing probability inference on a probability model.

  (Additional remark 9) The observation action determination means which determines the action of the robot which the distance measurement means can acquire distance information with high value when the said obstacle inference means performs efficient probability reasoning further in this computer system The autonomous movement program according to appendix 8, which is made to function as.

It is a top view which shows the general view of the autonomous mobile robot which concerns on one Embodiment of this invention. It is a figure which shows the functional structure of the autonomous mobile robot which concerns on this embodiment. It is a figure for demonstrating a Bayesian network, (a) shows an acyclic directed graph, (b), (c), and (d) show the probability table accompanying a node A, a node B, and a node C, respectively. . It is a figure which shows simply the structure of the Bayesian network as a probability model which the obstacle reasoning part of the autonomous mobile robot which concerns on this embodiment uses for a probability reasoning. It is a flowchart which shows operation | movement of the autonomous mobile robot which concerns on this embodiment.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Autonomous mobile robot 102a Left front distance sensor 102b Forward distance sensor 102c Right front distance sensor 202 Movement mechanism 204 Distance history evaluation part 206 Obstacle reasoning part 208 Observation behavior determination part 210 Autonomous movement control part 220 Obstacle

Claims (3)

An autonomous mobile robot that moves autonomously by recognizing obstacles and driving and controlling the movement mechanism,
A distance measuring means for measuring the distance to the obstacle;
Distance history evaluation means for evaluating the history of distance information measured by the distance measurement means;
Obstacle inference means for inferring an obstacle by performing probability inference on a probability model based on the behavior of the robot and the evaluation result by the distance history evaluation means;
Observation behavior determining means for determining the behavior of the robot that the distance measuring means can acquire high-value distance information for the obstacle reasoning means to perform efficient probability reasoning;
An autonomous mobile robot comprising: An obstacle identification method in an autonomous mobile robot that moves autonomously by driving and controlling a moving mechanism while recognizing an obstacle,
A distance measuring step in which the distance measuring means measures the distance to the obstacle; and
A distance history evaluation means for evaluating a history of distance information measured by the distance measurement step;
An obstacle inference step in which the obstacle inference means infers an obstacle by performing probability inference on a probability model based on the behavior of the robot and the evaluation result of the distance history evaluation step;
An observation behavior determination means for determining the behavior of the robot that can be acquired by the distance measurement means with high-value distance information when the obstacle reasoning means performs efficient probability reasoning; and
An obstacle identifying method for an autonomous mobile robot comprising: For autonomous mobile robots that move autonomously by driving and controlling the movement mechanism while recognizing obstacles,
A distance measuring means for measuring the distance to the obstacle;
Distance history evaluation means for evaluating the history of distance information measured by the distance measurement means;
Obstacle inference means for inferring an obstacle by performing probability inference on a probability model based on the behavior of the robot and the evaluation result by the distance history evaluation means;
Observation behavior determining means for determining the behavior of the robot that the distance measuring means can acquire high-value distance information for the obstacle reasoning means to perform efficient probability reasoning;
Program to function as.

Images