JPH0830327A - Active environment recognition system - Google Patents

Abstract

PURPOSE:To recognize a complicated environment by improving sensing capability by changing an external environment by generating a moving command to an actuator mechanism corresponding to information from a sensor mechanism for collecting the information of the external environment. CONSTITUTION:This system is provided with the hierarchical structure of a single or plural sensor mechanisms 1 for collecting the information of an external environment 3, actuator mechanism 2 for changing the external environment 3 and information processing mechanisms 4a-4c for generating the suitable moving command to the actuator mechanism 2 corresponding to the sensor information from the sensor mechanisms 1. Corresponding to the conditions at the sensing time of the external environment 3, the actuator mechanism 2 suitably changes the external environment 3 of itself and any object or the like so that the sensor mechanisms 1 can be sufficiently functioned. In this case, the low-order sensor information processing unit 4a repeats sampling and data processing in a short time, and the high-order sensor information processing unit 4b selects and commands the suitable operation of the entire system at comparatively long processing intervals while considering the final target of the system, etc., as well.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an active environment recognition system which recognizes an external environment around a sensor using various sensors.

In recent years, with the demand for automation of work processes in factories and unmanned maintenance work in nuclear plants, etc.,
There is an increasing demand for intelligent machines in which a robot or the like recognizes its surrounding environment by using various sensors such as a vision sensor and a proximity sensor and selects an appropriate action based on the environment. In particular, intelligent robots that have a visual sense of recognizing an external environment by using an optical vision sensor such as a TV camera or a range finder and adaptively perform complicated tasks are industrial robots based on conventional routine operations. Currently, research and development are being actively pursued as a next-generation robot that will replace the.

[0003]

2. Description of the Related Art FIG. 13 is a diagram for explaining a conventional environment recognition system. As shown in this figure, the basic idea in the conventional environment recognition system is to use the various sensors 1 in the environment recognition system that senses the external environment 3. From the fragmentary knowledge of environment 3, he tried to reconstruct the external environment as accurately as possible using information processing technology and various mathematical methods.

For example, in the research of environment recognition using vision, in order to recognize a three-dimensional environment around a robot from a two-dimensional image projected on a camera imaging surface by a television camera or the like, so far, David M
arr) 's famous book "Vision", "Computational methods for human image understanding and visual information processing", San Francisco, WH Friedman Company, 1982,
{A Computational Investigation into the Human Rep
retentation and Processing of Visual Information, S
An Francisco, WHFreedman and Company, 1982} has been proposed for many computational image understanding techniques. For example, in the field of structure reconstruction from disparity (shape from disparity) that creates a depth map from disparity between multiple images, represented by binocular stereoscopic vision (binocular streopisis),
Marr and Poggio's "Theory of Computation in Human Stereopsis" for the problem of determining the corresponding region between multiple images, London Royal Society Processing, B, Vol. 207, p. 187- 217, 1978, {"A computa
tional theory of human stereo vision ", Processingso
f the Royal Society of London, B, Vol.207, pp.187-21
7,1978,1979}, algorithms using zero crossings, and methods based on correlation operations based on the characteristics of regions have been proposed. These are occlusions {what the right eye sees cannot be seen by the left eye. Although there are still problems with "problem" and recognition of curved contours, relatively accurate results are obtained in general scenes.

In the shape from shading, which restores the three-dimensional shape of an object from the shadow of the surface of the object, a relaxation constraint method based on smooth constraint systematized as a standard regularization theory later is called Horn. "Structure Restoration from Shadows", Psychology Computer Vision, 1975,
{"Shape from shading", The Psychology of Computer
Vision, 1975} and Ikeuchi's "Reconstruction of mathematical structure from shadows and occlusion shape in monocular vision", AI Memo, 566, AI Laboratory, MIT, 1980, {"Numerical shape from shading
and occluding contours in a single view ", AI memo, N
o.566, AI Lab, MIT, 1980}.

Further, regarding "shape from motion" for detecting the relative depth from the motion of an image, the generation of optical flow by the relaxation method is performed by Horn and Schunk. "Optical Flow Determination", Artificial Intelligence, Volume 17, pp. 185-203, 1
981, {"Determining optical flow", Artificial Intell
igence, Vol17, pp185-203,1981} and Campan (Campan
i), Alessandro's "Dynamic analysis from primary generated optical flow", CVGIP, Volume 56, Volume 1
No., July 1992, {"Motion Analysis from First-Orde
r Properties of Optical Flow ", CVGIP, Vol.56, No.1, Jul
y 1992}, a method using the least squares method has been proposed.

[0007]

However, as described above, research on the conventional environment recognition system has been made on how to accurately reconstruct the external environment 3 with the ability of the sensor 1 to determine a given environment. The focus of the research was on.
Therefore, even if the various methods proposed to date are used, if the external environment 3 is too complicated, the sensor 1 itself has low capability, or the sensor 1 does not function sufficiently. In a poor environment, it is often difficult to accurately reconstruct the external environment 3, and a practical environment recognition system that operates in an actual environment has been rarely realized until now.

Looking at the example of computer vision, the past
For 35 years, in the framework of "reconstruction of 3D world from 2D vision", many important methods such as those mentioned above were proposed, but vision was used for real-time robot motion control. To date, few cases have been reported applied in the real environment. This is because the actual environment is so complicated and varied that even if various computer vision methods, environment description methods, high-performance visual sensors, and high-speed computers developed so far are used. This is because it was difficult to completely restore the external environment 3 in real time.

In view of the above-mentioned conventional drawbacks, it is an object of the present invention to provide a practical environment recognition system that operates in a real environment and an intelligent robot.

[0010]

1 to 3 are explanatory views of the principle of the present invention, and FIG. 1 shows an example of the overall configuration.
Shows the concept of the information processing unit, and FIG. 3 shows an example of the internal configuration of the information processing unit. The active environment recognition system according to the present invention does not use the complicated and diverse external environment 3 and its own sensing ability as they are, but instead of using the position of the sensor 1 itself or the object according to the situation at the time of sensing. By actively changing the self and external environment 3 by the actuator mechanism 2 such as a robot, such as position and orientation, lighting conditions, and removal of obstacles in some cases, it is possible to create a complex environment that was difficult with the conventional technology. It enables recognition.

That is, it is an active sensing system in which the interaction between the sensor system 1 and the external environment 3 is positively utilized in the recognition process to improve the sensing ability. Until now, such as active vision system
Research has been conducted to improve the sensing ability by changing the state of the sensor 1 side, but improving the sensing ability by changing the external environment 3 itself, which is the recognition target, such as movement of objects and removal of obstacles. The idea of trying to
This is a new concept that does not exist in the framework of conventional environment recognition systems.

The environment recognition system that positively utilizes such interaction with the external environment 3 has a flexible structure that can cope with the time constraints of processing for adaptively responding to environmental changes and the complexity of the environment. Therefore, the present invention uses the following hierarchical sensor information processing system as its basic configuration.

This is, as shown in FIG. 1, 1)
A direct actuator mechanism that does not consider the result of the action, such as spinal reflex, or "unconscious" action for humans, such as reflex action, from environmental information at the information level obtained directly from each sensor 1. Lower layer sensor information processing unit 4a which generates a motion command to the sensor, and 2) each sensor 1 and upper layer sensor information processing unit 4a.
c, information obtained by the lower layer sensor information processing unit 4a, and symbol-level environment information (for example, to represent a complicated object such as a keyboard, a key, a portion where the key is arranged, and the key are By using environment information that expresses a hierarchically structured object such as a housing composed of a plurality of parts with a predetermined symbol, logical inference with a low degree of abstraction, for example, recognizing an object hiding another object The information processing unit 4b for the middle layer, which generates a motion command to the actuator mechanism 2 by planning the local self-movement and the changing procedure of the external environment 3 by removing it, and 3) each sensor 1, and the lower layer. Using the information obtained by the sensor information processing unit 4a and the above-mentioned symbol-level environmental information, well-known Bayesian inference {In summary, multiple factors interact with each other in external environment 3. One of these factors is taken out, and a predetermined action, for example, moving in a certain direction, looking, or touching, causes other factors to change, and the external environment 3 is a reasoning mechanism that infers what happens to 3 and determines the most appropriate way to move it.} By using a logical reasoning mechanism with a high degree of abstraction, the final goal of the active environment recognition system and the sensor mechanism 1 are Global sensory movement for fully functioning, upper layer sensor information processing unit 4c that infers and selects the procedure of changing external environment 3 and generates a motion command to actuator mechanism 2. This is realized by stacking the information processing units 4a, 4b and 4c in a hierarchical manner as shown in FIG.

Of these, the lower layer sensor information processing unit 4a described in the above section 1) repeats sampling (data acquisition) and data processing in a short time, and the basic operation between the actuator 2-environment such as contact and collision. It monitors and controls various interactions.

The upper layer sensor information processing unit 4c described in the above section 3) also takes into consideration the final goal of the system such as object grasping at relatively long processing intervals, and the efficiency of sensing and environmental changes. Select and command the appropriate movement of the entire system.

Further, a mechanism 40 for exchanging the recognized information about the external environment 3 and fusing them to perform inference with high overall consistency is provided between these layers. For example, when the hand of the robot touches the cup, the above-mentioned middle layer sensor information processing unit 4b is
The information from the lower layer sensor information processing unit 4a that "contacted with some object" and the upper layer sensor information processing unit that "there should be a cup here"
It takes the information from 4c and corrects the "nothing in there" or "this is a bottle" reasoning itself.

Further, each layer includes, for example, information about the external environment 3 by the active environment recognition system according to the present invention,
If the reasoning device 40 such as the Bayesian reasoning described above does not select an appropriate self-movement or a changing procedure of the external environment 3, for example, a trial-and-error method using random numbers, etc. It has a mechanism to avoid falling into a deadlock by determining the next action in Method 42.

That is, for example, when the upper layer cannot recognize the external environment 3 at all, the upper layer informs the lower layer that "the position of the sensor or the object in the environment should be appropriately moved" or "the lighting condition is changed." It is said that it will create an environment that is easy to recognize by trial and error by sending exercise commands that change the external environment 3 such as self-movement.

In each layer, a mechanism for learning and acquiring environmental recognition based on sensor information and self-movement and environmental change procedures realized by the inference method 40 and the trial-and-error method 42, such as a neural network ( There is a learning mechanism).

This learning mechanism includes an environment recognition mechanism 45, and a learning mechanism 44 that directly generates a motion command from the sensor information and a conventional environment recognition mechanism 45 are used to extract the environment information from the sensor information. , A learning mechanism 43 that generates a motor command based on it.

Generally, a learning mechanism including an environment recognition mechanism 44
Is a learning mechanism that can generate low-level motor commands such as reflexes at high speed and does not include the environment recognition mechanism.
3 takes time to recognize the environment by the environment recognition mechanism 45, but can generate a high-level motion command such as a typical voluntary motion at longer time intervals based on the recognized complicated environment information. .

By these learning mechanisms 43 and 44, it is possible to improve system performance such as reduction of inference cost and improvement of robustness (disturbance resistance).

[0023]

Therefore, the present environment recognition system can recognize the complicated external environment 3 which is difficult to be accurately recognized by the conventional environment recognition system, by the given sensor ability. Since the self and external environment 3 are positively changed, a practical environment recognition system that operates in the real environment can be constructed.

As an example, consider a situation in which a part of an object is hidden by another object in a visual environment recognition system. The framework of robot vision research so far, represented by Marr mentioned above, looks at these objects from a certain point in space, and recognizes each object by various computer vision methods. For example, we try to reconstruct the three-dimensional environment in the visual field as much as possible by performing object identification and position and orientation estimation.

Therefore, in the case where the concealment state is very complicated or the lighting condition is poor, there is a possibility that the conventional computer vision technique cannot recognize the object.

On the other hand, the basic idea of the present invention is that if an object cannot be recognized only by looking at it from a certain point, the viewpoint is moved to a position where it can be seen better, or the object is hidden in a complicated manner for recognition. For difficult objects, the visual system such as the sensor 1 can easily function by positively changing the self and the external environment 3, such as moving the obstacle to a place that does not hide the obstacle or changing the lighting conditions appropriately. It is said that the environment is created by itself.

Therefore, in the conventional visual recognition system or mobile robot system, even if it is difficult to achieve the target due to the sensor capability and the complexity of the environment, the environment is positively organized to finally recognize the external environment 3. However, it is possible to realize a sensor information processing system that enables achievement of the target, and it is possible to obtain an effect that it greatly contributes to the performance improvement of the environment recognition system.

[0028]

Embodiments of the present invention will be described in detail below with reference to the drawings. 1 to 3 described above are explanatory views of the principle of the present invention.
4 to 7 are views showing an embodiment of the present invention,
4 shows an outline of an object recognition / grasping system by a television camera, FIGS. 5 to 7 show configuration examples of a hierarchical sensor information processing system, and FIGS. 8 to 12 show another embodiment of the present invention. FIG. 8 shows an image of a group mobile robot system, FIG. 9 shows an example of a group moving method of the group mobile robot, FIG. 10 shows an example of an obstacle, and FIG.
0 shows a flow chart of a process of moving an obstacle in the group mobile robot, and FIG. 11 shows a process of moving an obstacle in the group mobile robot as a concrete example.

In the present invention, a single or a plurality of sensor mechanisms 1 for collecting information on the external environment 3 and the external environment 3 are used.
Actuator mechanism 2 that changes the
From the sensor information from 1, the above actuator mechanism
Information processing mechanism 4a that generates an appropriate motion command to 2
4c and 4c in a hierarchical structure, the actuator mechanism 2 operates in accordance with the sensing conditions of the external environment 3 so that the sensor mechanism 1 fully functions and the external environment 3 Is a means necessary for carrying out the present invention. Note that the same reference numerals indicate the same object throughout the drawings.

Hereinafter, referring to FIGS. 1 to 3, FIGS.
The configuration and operation of the active environment recognition system of the present invention will be described with reference to FIGS. 7 and 8 to 12.
In one embodiment of the present invention, FIGS. 4 to 7 show an example of an object recognition and gripping system by a television camera, and in another embodiment, FIGS. 8 to 12 show a group mobile robot system.

First, the active environment recognition system of the present invention will be described by taking the object recognition and grip system by the television camera shown in FIG. 4 as an example. This system consists of a 6-axis vertical articulated robot and controller, a TV camera attached to the robot hand, a force sensor, a tactile sensor, a proximity sensor, and a computer for realizing an image processing operation and a hierarchical information processing system. Become.

The environmental information obtained from the television camera 1a includes, for example, the movement (u _i, v _i ) of the object image on the image plane calculated by the following constraint condition expression. I _xi u _i + I _yi v _i + I _ti = 0 where I _xi , I _yi , and I _ti are points i on the screen, respectively.
It is a change amount of the pixel gray value I _i in (x _i, y _i ) with respect to the x and y directions and time t. This (u _i, v _i )
Is generally called an optical flow, and the distance to the object in the screen and the moving speed of the object are calculated from this and the speed of the robot hand measured by the internal sensor to detect the possibility of collision with the hand. An object can be detected at high speed.

In addition to this, information of an object image cut out by color and gray value information, template matching,
_, Geometric model (so-called, CAD model) types of gaze object identified by matching with the position, such as information of posture, is input to each layer as the environment information obtained from the television camera 1a respectively.

On the other hand, as the sensor information other than that of the TV camera 1a, the acting force on the robot hand obtained from the force sensor 1b, the contact state with the environment by the tactile sensor 1c, the relative distance information with the environment by the proximity sensor 1d, etc. is there.

Next, a detailed example of the hierarchical sensor information processing system shown in FIG. 4 will be described with reference to FIGS. As shown in the figure, the lower layer sensor information processing unit 4a uses the optical flow obtained from the TV camera 1a and the proximity sensor 1d to quickly avoid collision with the external environment 3 and make contact with it. Control and force sensor
1b, The output of the tactile sensor 1c is used to construct an impedance control system {a control system that performs operations such as smooth stroking and touching}, and controls the basic interaction between the hand and the external environment 3.

The intermediate layer sensor information processing unit shown in FIG.
4b identifies the type, position, and orientation of the gaze object by clipping the object by color / gray values, template matching, and matching the object image with the geometric model of the system, and performs local recognition to grasp the object. Make a proper work plan. However, in this work plan, for example, when the expected grip point is on the back side of the object, the object is pushed down to bring the grip point closer, or when it is determined that the object cannot be identified. , Behaviors that positively utilize interaction with the external environment 3, such as changing the position and orientation of an object by trial and error using random numbers, moving an obstacle that is blocking, and so on.

In the upper layer sensor information processing unit 4c shown in FIG. 7, the environmental information (such as which object was placed at which position) and the sensor information obtained by the lower layer sensor information processing unit 4a are used. Color of the entire field of view obtained
In order to efficiently move to the position (work target) of the target gripping target, or the target object, from the positional relationship between the objects that the system has in advance, based on the gray value image, using a predetermined inference mechanism 40, etc. Creates a comprehensive work plan that considers (inference, prediction) which obstacles to move including the gaze object (work efficiency).

This global work plan is created at a relatively long time interval after the environmental information is acquired to some extent, as compared with the lowest collision avoidance unit. Also, if it is determined that the current environment is complicated and unrecognizable, a lower unit (middle layer,
Attempts to create a recognizable environment by instructing the lower layer sensor information processing units 4b, 4a etc.) to move a grippable object.

Next, an example of environment recognition in the group mobile robot system using the present invention will be described with reference to FIGS. This system, as shown in FIG. 8, includes a manipulator and various sensors such as a television camera and a distance sensor.
It is composed of a plurality of robots 2a equipped with, and each robot 2a can freely move in the outdoor environment and can create an environment map and perform various works.

In this system, when determining where each robot 2a is now or how to move, in addition to the information from the sensor mounted on each robot 2a, the robots 2a other than itself and the environment Use a specific object inside as a landmark.

An example of the moving method of this system is shown in FIG. For example, if the robot (3) 2a moves, the positions of the other robots (2) and (3) 2a viewed from the robot (1) 2a are measured and the relative angles θ ₁ and φ ₁ from the reference position are measured. The robot (2) 2a recognizes the positions of the other robots (1) and (3) 2a by measuring the relative angles θ _{2 and} φ ₂ from the reference position _, and then recognizes the mutual position information. Replace, for example, robot (1)
2a estimates the position of the robot (3) and knows its own moving direction.

Further, when an obstacle 3a exists between the plurality of robots (1), (2) 2a, and the position of another robot (2) 2a cannot be recognized from the robot (1) 2a. The environmental arrangement according to the present invention, specifically, the movement of the obstacle 3a is realized as follows.

That is, in this system, as shown in FIG. 10, when the other robot (2) 2a or a mark in the environment cannot be observed due to an obstacle 3a or the like, it is difficult to identify the self-position and determine the movement strategy. In this case, as shown in FIG. 11 and FIG. 12, not only the robot itself is moved as in the conventional case, but also other robots are instructed to move, obstacles are moved, and the environment is positively organized. By trying to solve the problem.

That is, FIG. 11 is a flow chart showing an operation example in the case of recognizing another robot 2a in the group mobile robot system. First, the self moves as shown in FIG. 12 (a). {Processing step of FIG. 11
See 100} Find the other robot 2a that is the target, and if the other robot 2a cannot be recognized, try moving the other robot 2a as shown in FIG. 12 (b). {Processing step of FIG. 11
See 101, 102, 104} In this way, when the other robot 2a can be recognized,
Identify your location. {Processing steps 102,1 in FIG. 11
04, 101, 102, 106} In the above processing step 102, when the other robot 2a cannot be recognized, an attempt is made to move the obstacle 3a as shown in FIG. 12 (c). {Refer to processing steps 102 and 105 in FIG. 11} When the position of the self can be identified, it is checked whether or not the destination is reached. If not, the process returns to the processing step 100 to repeat the same operation. {Processing step 1 of FIG. 10
Also, this system has hierarchical sensor information processing units 4a and 4c (see FIGS. 5 to 7) composed of upper and lower two layers. Lower layer sensor information processing unit 4a
It controls the basic functions required for autonomous behavior of the robot 2a, such as recognition of a and observation of other robots 2a and landmarks in the environment.

The upper layer sensor information processing unit 2c
Generates a global command for achieving a target such as movement of the obstacle 3a and generation of a motion command to another robot 2a. In this way, using this system, even if the object is hidden by the obstacle 3a and the environment is complicated and accurate recognition is difficult with the conventional image processing system, the obstacle 3a is removed. By adjusting the environment by yourself, it becomes possible to finally carry out the recognition work.

As described above, in the active environment recognition system according to the present invention, when it is difficult to accurately recognize the external environment 3, the function of appropriately changing oneself or the environment according to the situation, and the accurate recognition of the external environment. If it is possible, the function of reconstructing the external environment 3 from the sensor information and creating an advanced work plan, and if the external environment 3 is complicated and difficult to recognize accurately, the recognition system first Attempting to organize the environment by using the lower layer sensor information processing unit 4a, the external environment 3 was organized to some extent, and the upper layer sensor processing unit 4b, ~ became available for more accurate environment recognition. The work plan of is created and the high-level recognition work is performed.

That is, the lower layer sensor information processing unit 4a here is a spinal reflex, or "unconscious" behavior in humans, and the lower layer sensor information processing unit 4b,
~ Is a "conscious" action based on knowledge and experience.
Then, in the present invention, when using this hierarchical information processing system, for example, when it is desired to pick up an invisible object obstructed by the obstacle 3a by hand, it is not necessary to observe the obstacle 3a in detail. The sensor information processing unit 4a unintentionally extends the hand to move the obstacle 3a, and when the object is observed, the upper layer sensor information processing unit 4a
b, ... are used to accurately recognize their type, position, and posture.

Therefore, when the object image to be recognized is sufficiently clear and the object is isolated from the surrounding environment,
That is, when the target can be recognized even by the conventional image processing method, the target is identified by using the upper layer sensor information processing system 4b, to further recognize the position, the posture, etc. of the target more accurately. Change the position of sensor 1. If the object is hidden by the obstacle 3a and the recognition is difficult with the conventional image processing method such as when the object is hidden by the obstacle 3a, first use the lower layer sensor information processing unit 4a.
The sensor system functions better by approaching 3a and moving by grasping the obstacle 3a that shields the target appropriately, or by moving the object itself to be recognized to a place without the obstacle 3a. The external environment 3 can be organized so that it can be easily recognized, and then the object can be recognized using the upper layer sensor information processing units 4b ,.

[0049]

As described above in detail, according to the present invention, in the conventional visual recognition system and mobile robot system, even if it is difficult to achieve the target due to the sensor capability and the complexity of the environment, By organizing the environment into
In the end, a sensor information processing system that recognizes the external environment and achieves the target can be realized, and it greatly contributes to the performance improvement of the environment recognition system.

[Brief description of drawings]

FIG. 1 is an explanatory diagram (1) of the principle of the present invention.

FIG. 2 is an explanatory diagram of the principle of the present invention (No. 2)

FIG. 3 is an explanatory diagram of the principle of the present invention (No. 3)

FIG. 4 is a diagram showing an embodiment of the present invention (No. 1).

FIG. 5 is a diagram showing an embodiment of the present invention (part 2).

FIG. 6 is a diagram showing an embodiment of the present invention (part 3).

FIG. 7 is a diagram showing an embodiment of the present invention (Part 4).

FIG. 8 is a diagram showing another embodiment of the present invention (No. 1).

FIG. 9 is a diagram showing another embodiment of the present invention (Part 2).

FIG. 10 is a diagram showing another embodiment of the present invention (part 3).

FIG. 11 is a diagram showing another embodiment of the present invention (Part 4).

FIG. 12 is a diagram showing another embodiment of the present invention (No. 5).

FIG. 13 is a diagram illustrating a conventional environment recognition system.

[Explanation of symbols]

1 Sensor mechanism 2 Actuator mechanism 3 External environment 4 Sensor information processing unit 4a Lower layer sensor information processing unit 4b Middle layer sensor information processing unit 4c Upper layer sensor information processing unit 2a Robot 3a Obstacle 40 Inference mechanism 41 Switching means 42 Trial and error Decision mechanism 43,44 Learning mechanism 45 Environment recognition mechanism 100-107 Processing steps Sensor information Motion command Lower layer sensor Information from information processing unit

Claims (8)

[Claims] 1. A single or plural sensor mechanism (1) for collecting information of an external environment (3), an actuator mechanism (2) for changing the external environment (3), and a sensor mechanism (1) from the sensor mechanism (1). Based on the sensor information (), the above actuator mechanism (2)
An active environment recognition system characterized by comprising an information processing mechanism (4) for generating a motion command () to the user. 2. The active environment recognition system according to claim 1, further comprising a predetermined inference mechanism (40) in the information processing mechanism (4) of the active environment recognition system, (4) uses the information about the external environment (3) obtained from the sensor mechanism (1), and the reasoning mechanism (40) in the information processing mechanism (4) detects the self-motion, the position of the object, An active environment recognition system characterized by generating and issuing a motion command () to the actuator mechanism (2) by inferring and selecting a procedure for changing the external environment such as lighting. 3. The active environment recognition system according to claim 1, wherein the information processing mechanism (4) of the active environment recognition system comprises:
An active environment recognition system comprising a trial-and-error decision mechanism (42) for determining the next action by a trial-and-error method. 4. The active environment recognition system according to claim 1, further comprising a learning mechanism (43, 44) in the information processing mechanism (4) of the active environment recognition system, (4) The self-movement and the external environment change procedure realized by the inference mechanism (40) in (4) and the trial-and-error decision mechanism (42) that determines the next action by a trial-and-error method are described above. 43,44) is an active environment recognition system characterized by learning and acquisition. 5. The active environment recognition system according to claim 4, wherein the learning device (43, 44) is a neural network. 6. The active environment recognition system according to claim 1, wherein an information processing mechanism for generating a motion command to the actuator mechanism (2) based on sensor information from the sensor mechanism (1). 4) is an information processing mechanism (4a) that generates a direct motion command based on the sensor information from the sensor mechanism (1), information () from the sensor mechanism (1), and Using the information (), etc. obtained from the information processing mechanism (4a) etc. that generates various movement commands, the predetermined inference by the above inference mechanism (40) and the change of the external environment (3) An active environment recognition system characterized by being a hierarchical information processing system composed of an information processing mechanism (4b, ...) which generates a motion command () for the actuator mechanism (2) by planning a procedure. . 7. The active environment recognition system according to claim 1, wherein one of the sensor mechanisms (1) includes a vision sensor (1a), and the self obtained from the vision sensor (1a), Also, as an environment recognition means (45) for the self and the external environment (3) by the information of the external environment (3), an object detection means using an optical flow, a means for cutting out the target object by the color and gray value information, and a template An active environment recognition system characterized by using means for recognizing types, positions, and orientations of objects by matching and matching of geometric models of object images. 8. A group robot position identification system in which a plurality of robot mechanisms (2a) cooperate with each other to identify positions of each other, and an obstacle (3a) or the like recognizes another robot mechanism (2a). If it is difficult to move, move yourself or another robot mechanism (2a) or move the obstacle (3a) to the opponent robot mechanism (2a).
An active environment recognition system characterized by recognizing objects and identifying positions.