WO2015063119A1

WO2015063119A1 - Autonomous multi-modal neuro-inspired mobile robot for monitoring and restoring an environment

Info

Publication number: WO2015063119A1
Application number: PCT/EP2014/073168
Authority: WO
Inventors: Ramesh CAUSSY; Pierre DELARBOULAS
Original assignee: Partnering 3.0
Priority date: 2013-09-12
Filing date: 2014-10-29
Publication date: 2015-05-07
Also published as: FR3010529A1; FR3010528A1; FR3010529B1

Abstract

The invention relates to a mobile robot (100) comprises movement and orientation means (110) and sensory means, comprising: a. sensory means able to deliver proprioceptive information from the movement and orientation means (110), b. sensory means (130, 141, 142, 143, 144, 145, 146, 147) able to deliver so-called environmental information perceived in the robot's environment; characterised in that it comprises computer means comprising: c. a neuro-mimetic structure comprising a battery of idiothetic grid cells whereof the neurons are activated by proprioceptive information and a battery of allothetic grid cells; d. a neuro-mimetic structure, called entorhinal cortex, each neuron of which corresponds to a place cell.

Description

AUTONOMOUS MULTI-MODEL NEURO-INSPIRED MOBILE ROBOT FOR MONITORING AND RECOVERING AN ENVIRONMENT

The invention relates to a mobile robot for autonomous monitoring of an environment. The invention is more particularly, but not exclusively, intended for the field of environmental monitoring including the monitoring of air quality in a closed environment.

In the field of monitoring the atmospheric state of a closed environment, it is known to use different detection devices within this enclosed environment. A detection device generally makes it possible to detect a quality parameter of the atmospheric state in a single room of a room, each detection device ensuring the detection of a single environmental status parameter. For example, it is necessary to place, in each room of the room, a smoke detector to detect possible departures of fires. It is also known to install pollution detectors for detecting gases or toxic products in factories and for detecting viruses, bacteria and other allergens in medical premises. It is of course conventional to install thermometers and humidity sensors to know the ambient temperature and the humidity of a room.

However, with these systems, it is necessary to install several detectors distributed in the different rooms or different locations of the enclosed environment. To monitor several environmental status parameters, the installation of the many detectors needed to monitor the entire enclosed environment is therefore relatively expensive.

In addition, these systems have the disadvantage of giving the value of an environmental status parameter at a given location, while a few meters further, the parameter is likely to take a different value. This is the case, for example, of an ambient temperature which may vary from one location to another of the same room depending on the orientation of the room relative to the sun or depending on the presence of ducts. heating, air conditioning ducts, etc. Similarly, some areas may have a different humidity for reasons similar to those explained previously. Thus, for appropriate monitoring of the environment in a closed room, it is advantageous to use one or more surveillance robots, like security agents performing regular rounds. The automation of these tasks, however, faces complex problems. Indeed, a security officer is able to instantly assimilate the round trip to perform, not only in terms of travel but also places to visit. It is able to significantly modify its course according to the evolution of the configuration of the places or the priorities, while ensuring the effective control of the places of which it must ensure the security.

The document "An Intelligent Environmental Monitoring System Based on Mobile Mobile Robots" by JUNJUN WU et al., IEEE International Conference on Robotics and Biomimetics, (ROBIO 201 1) - December 7, 201 1, describes such a system.

This document describes the deployment of a robotic solution in industrial environments, similar to large warehouses, whose size is simple, stable and predictable. The robot is autonomous and localized in the environment according to a simultaneous mapping and localization method known by the acronym SLAM or Simultaneous Location And Mapping. According to this method, the robot constructs a cartesian cartography of the places by means of its sensors, and stores it in memory means. It is then identified in this map. The construction of the cartography is carried out by means of the sensors of the robot, in particular by stereoscopic vision systems making it possible to estimate distances or laser scanning systems. If the environment evolves, the robot must rebuild its mapping. From a practical point of view, this method requires the recording of an initial mapping and the progressive enrichment of it, so that when the environment is changing, the robot spends more time mapping than performing monitoring tasks, and that memory requirements increase as mapping becomes more complex.

The invention aims to solve the disadvantages of the prior art and concerns for this purpose a mobile robot provided with sensory means comprising:

at. sensory means capable of delivering proprioceptive information from the moving and orienting means;

b. sensory means capable of delivering information, called environmental information, perceived in the environment of the robot; characterized in that it comprises computer means comprising:

vs. a neuromimetic structure comprising a battery of idiothetic gate cells whose neurons are activated by proprioceptive information and a battery of allothetic gate cells;

d. a neuromimetic structure, called entorhinal cortex, where each neuron corresponds to a place cell.

Thus the robot is able to learn and recognize its environment by the information delivered by its sensory means without using a map but only its cognitive abilities. The robot object of the invention learns and understands an environment that emerges, completely autonomously without acquiring or updating a map. The density of the neuromimetic structures is fixed, the overall number of neurons does not change, so that the learning and recognition of new places or the self-tracking of the robot in a changing environment, do not change the amount of memory needed . The robot object of the invention combines for its navigation proprioceptive sensory information, or idiothetic sensory information and environmental or allothetic, both types of information being acquired continuously during the exploration of its environment.

The invention is advantageously implemented according to the embodiments described below, which are to be considered individually or in any technically operative combination.

Advantageously, the robot object of the invention comprises:

e. computer means for merging the information of the allothetic gate cell battery and the idiothetic gate cell array.

Thus the robot object of the invention is able to locate even in the absence of one or more allothetic information which allows it to move in a changing space without the need to map this space.

Advantageously, the robot object of the invention comprises:

f. a neuromimetic structure, called PrPh, whose neurons are activated by the combination of proprioceptive information and environmental information. Thus, this matrix structure makes it possible to associate a configuration of the environment's allothetic bitters with a position of the robot and to predict the position of the robot, as a function of the visual bitters picked up by it, whatever its later position vis-à-vis these bitter ones.

According to an exemplary embodiment, the proprioceptive sensory means comprise one or more of:

- an odometer;

an orientation sensor;

- a speed sensor.

The combination of all or part of this information enables the robot to find its way in space through a procedural memorization and spatial recognition process, and to navigate in the blind environment or with partial visibility of the locating information.

According to an exemplary embodiment, the environmental sensory means comprise one or more of:

- a vision sensor;

- a dust detector;

- a humidity detector;

- a toxic product detector;

- a gas detector;

- a smoke detector;

- a microphone.

Thus the robot object of the invention is capable of constructing and recognizing a multisensory allothetic image of the environment. This image is used for auto-tracking the environment but also to contextualize measurements made in this environment for monitoring purposes.

According to a particular embodiment of the robot object of the invention, the vision sensor is a panoramic camera and includes proprioceptive means for measuring the orientation of the vision of said camera. Thus the combination of the visual information with the information of the orientation of the camera makes it possible to recruit cells in the PrPh.

Advantageously, the robot object of the invention comprises means, said digital compass, able to determine the position of the robot with respect to an allothetic mark. Thus the robot object of the invention and capable at least locally to define its position absolutely in the environment in which it evolves.

Advantageously, the robot object of the invention comprises a plurality of proximity sensors capable of preventing a collision.

According to a particular embodiment, the robot object of the invention comprises means, said effectors, able to modify the environment of the robot. Thus, in addition to performing environmental monitoring, the robot object of the invention is able to act on said environment to address the risks as soon as they are detected.

Advantageously, said effector means comprise:

- a fragrance diffuser; or

a chemical or biological depollution reactor; or

- an air humidifier.

Advantageously, the robot object of the invention comprises alarm means. The invention also relates to a method for navigating in a closed environment by means of any of the embodiments of the invention, which method comprises the learning steps of:

i. recruit a place cell neuron into the entorhinal cortex;

ii. recruit environmental clues using the environmental sensor and merge these clues with their position in the robot environment by recruiting neurons into the PrPh; iii. associate with this configuration of the PrPh the configuration of the grid cells;

so that the neuron recruited in the entorhinal cortex reacts to the configuration of the fused allothetic and idiothetic grid cells corresponding to the learned configuration.

This mechanism allows learning the robot and allows it to adapt to any change in the environment.

Advantageously, the method which is the subject of the invention comprises the steps of autonomous navigation consisting of:

iv. fuse grid grid idiothetic grid cells derived information from proprioceptive sensory means and cells grids derived from PrPh forecasts;

vi. determine the cell of the most active place of the entorhinal cortex as a function of the grid cells resulting from the fusion;

vii. to deduce the position of the robot and the sequence of actions.

Thus, the navigation system of the robot that is the subject of the invention permanently corrects the errors of cumulating the idiothetic information but also uses this idiothetic information to compensate for the unavailability of certain allothetic information. This mechanism also allows the robot to recognize the location where it is even if the path to reach this place has been changed, for example, following a change in the configuration of the environment.

According to an advantageous embodiment of the method which is the subject of the invention, implementing a robot comprising a plurality of environmental sensors, the environmental indices recruited in step ii) are visual bitters and the method comprises the steps of:

viii. recruiting environmental indices using other environmental sensors;

ix associate with the configuration of the environmental indicators recruited in step viii) the configuration of the grid cells.

Thus the environmental information is contextualized in relation to the place where the robot is located. The use of visual amers, allows the robot to precisely define its position in the environment.

According to this last embodiment, the method which is the subject of the invention advantageously comprises a step consisting of:

x. categorizing the environmental indices recruited in step viii) and that the grid cell association of step ix) is performed by category.

Contextualization can detect combinations of potentially undesirable environmental parameters.

Advantageously, the method of the invention comprises a navigation step in which the robot enters an unexplored location and which comprises steps i) to iii) so as to integrate this place in its memory.

According to an advantageous embodiment of the method which is the subject of the invention, it comprises the steps of: xi. if the information coming from a sensory means of environment deviates significantly from the known level in this place;

xii. act on the environment by means of an effector or trigger an alarm.

Thus, the robot adapts its surveillance and its modes of action to the context of the places visited.

According to a particular embodiment, the method which is the subject of the invention comprises a step of

xiii. moving the robot into the environment following the gradient of the information detected in step viii) to go back to the source of the emission.

Thus the robot object of the invention is able to use its cognitive abilities to locate a source outside of its usual routes.

The invention is explained below according to its preferred embodiments, in no way limiting, and with reference to FIGS. 1 to 4 in which:

- Figure 1 is a front view of an embodiment of the robot object of the invention, Figure 1A according to an external view, Figure 1 B in internal view;

FIG. 2 is a schematic view of an exemplary embodiment of the neuromimetic control structure of the robot that is the subject of the invention;

FIG. 3 illustrates in a schematic diagram the neuromimetic structure for learning, memorizing and recognizing the idiothetic indices of a robot according to the invention;

and FIG. 4 shows, in a schematic representation, an exemplary embodiment of the neuromimetic structure for learning, memorizing and recognizing the allothetic indices of the robot that is the subject of the invention.

The mobile robot object of the invention is a robot dedicated to the control of the environment in a closed environment to ensure the safety of goods and people in said environment and suitable conditions of comfort. According to non-limiting application examples, said enclosed environment is a residential or office building, a company, a factory, or any other place delimited in the space requiring a monitoring of its interior environment.

According to an exemplary embodiment, the robot object of the invention performs the tasks a vigil by monitoring the premises and generating alarms or acting when anomalies are found.

The navigation system of the robot object of the invention is based on a principle of memorization by learning inspired neuroscience. Thus the robot object of the invention recognizes and continuously learns its environment according to a dual process of declarative spatial storage, that is to say the spatiotemporal storage of landmarks or bitter of its environment, and procedural spatial storage, allowing him to acquire a particular motor behavior to achieve his goal. These two processes complement each other continuously. Thus, the spatial memory of the robot object of the invention allows it to explore and learn its environment completely independently, even in an initially unknown environment, or more effectively, to change its memory environment from an initial learning, for example the monitoring path to achieve. The robot uses for this purpose a neuromimetic computer system in the form of neural networks. These sets of neurons mimic the functioning of the cells included in the hippocampus, the entorhinal cortex, the set, called PrPh, of the perirhinal cortex and the parahippocampal cortex of the mammalian brain. These areas of the brain include so-called place cells and so-called grid cells.

The place cells are the support of the cognitive map. Each neuron corresponds to a precise position of the robot and its activity is maximum when the robot is in this position. The place cells represent the declarative spatial memory.

The grid cells are sets of neurons with multiple activity peaks depending on the position of the robot and thus represent a procedural memory of the movement of the robot integrating its path.

According to the invention, these storage processes are performed with a constant amount of memory, represented by the number of neurons included in the neural networks implemented in the robot's computer navigation system, the learning being carried out by modifying the weights synaptic connections between neurons.

1, according to an exemplary embodiment, the robot (100) object of the invention comprises motor means, for example a pair of wheels (1 10) motor driven by electric geared motors (11 1). Each geared motor includes an encoder for measuring the angular displacement of the wheels (1 10) and their rotational speed. According to this exemplary embodiment, when said wheels (110) rotate in the same direction and at the same speed the robot advances in a rectilinear trajectory, the relative speed variation between the two drive wheels makes it possible to follow a curved trajectory, a function of the composition of speeds. When said wheels (110) rotate in opposite directions, the robot (100) rotates about its axis (101). According to this exemplary embodiment, the robot (100) comprises vision means (130) comprising a panoramic camera mounted on a rotating shaft (131) whose orientation relative to the robot is also measured by an encoder. This camera is an environmental sensor and allows the robot to recruit allothetic information in its environment. The encoders installed on the driving wheels and the encoder of the orientation shaft of the panoramic camera (130) constitute, according to this example embodiment, proprioceptive sensory means of the robot. According to this embodiment, the encoders installed on the driving wheels are used to recruit idiothetic information on the position of the robot in its environment, the information of the encoder on the shaft (31) of the orientation of the camera (130) panoramic are used to enrich the allothetic vision information. According to this exemplary embodiment, the robot that is the subject of the invention comprises other sensors enabling it to apprehend its environment. Thus, the lower part of the robot (100) comprises a town gas sensor (141), a humidity sensor (143), a pollutant gas sensor (144), for example an ozone sensor, and a sensor. dust (142). The front high part of the robot object of the invention comprises, for example, a toxic product detector (145), a thermometer (146) and a smoke detector (147) without this list being exhaustive. Advantageously, the robot also comprises a microphone (not shown) or a sound level meter enabling it to measure a sound environment or to detect a sound signal, and, in the intended application, these complementary environmental sensors are used as detectors, for example for detect a gas leak, or are also used to obtain allothetic information about the environment and supplement the vision information.

These sensors and detectors are positioned on the robot according to the mode of diffusion and propagation of the environmental information to be detected or measured. For example, the detections concerning the presence of heavier products than air is advantageously detected by sensors or detectors located at the bottom of the robot (100). On the contrary, the detections concerning the presence of volatile products are advantageously carried out by detectors or sensors located in the upper part of the robot (100). According to a particular embodiment, the robot which is the subject of the invention also comprises means for acting on the environment, for example a UV lamp reactor (180) capable of destroying allergens, certain pollutants such as formaldehyde or benzene as well as bacteria, or a fragrance diffuser (170). The air treated by the reactor (180) enters through peripheral openings (not shown) made in the upper part of the robot, then through said reactor (180) where it is purified and finally evacuated through orifices (not shown) located at the bottom of the robot. According to an exemplary embodiment, the robot (100) which is the subject of the invention comprises means for communicating, in particular light means (191), for example in the form of a flash, and a loudspeaker (192) for generating signals. alarm. A plurality of proximity sensors (160), for example, ultrasonic transducers are installed on the periphery of the robot and enable it to detect a risk of collision with an obstacle. All of the functions of the robot are controlled by an on-board computer (150) that includes the neuromimetic system for autonomous navigation of the robot. This on-board computer (150) communicates in particular by means of a display screen (151), and via a wireless network with a base.

Figure 2, the neuromimetic system of the robot object of the invention comprises 3 modules (210, 220, 230) neuromimetics. A first module (210) reproduces a cognitive functioning similar to the perirhinal and parahippocampal cortex and allows to memorize, recognize and predict a grid cell configuration (201) from allothetic information combining sensory information on the environment, for example vision, and proprioceptive information, for example the orientation of the panoramic camera. A second module (220) is capable of storing and recognizing grid cells (202) from idiothetic information from proprioceptive sensors, for example wheel speeds of the robot. The third module (230), whose operation is similar to that of the hippocampal operation and the entorhinal cortex, makes it possible to memorize and to recognize place cells from allothetic information or from a grid cell configuration. A fourth module (240) performs a synthesis between the grid cell patterns (202) recognized from the purely idiothetic information in the second module (220) and the grid cells (201) recognized from the allothetic information in the first module (210). This fourth module (240) makes it possible to correct the drifts generated by the cumulative errors of the purely idiothetic locating, but also to the robot to orient itself in conditions of poor visibility when the allothetic indices are not easy to discern.

The software components are built using neural networks. The neurons are grouped into groups or neuronal fields each representing a set of neurons following the same rules of learning and activation.

According to a preferred embodiment, the robot is learned from a location allowing the robot to connect to an energy network to recharge its batteries, so that the robot is always able to return to this location in due time.

Figure 3, in learning phase, the robot moves in the environment and discovers the places. This learning is faster if the robot is guided, but the robot is able to perform this learning task autonomously. During this phase, the second module (220) acquires the unit movements (310) of the robot (speed, direction) from the proprioceptive information from the encoders placed on the motricity means of the robot. Unit motion is used to generate grid cells without the need for a Cartesian map. The activity of a pair of neurons (301 1, 3012, 3013, 3014) of the randomly chosen neuronal field (301) is summed and then discretized on other fields (302). These fields are then compressed by modulo projection on smaller fields (303). The conjunction of two of these fields is sufficient to obtain a matrix (304) of grid cells.

Figure 4, still in the learning phase, when the robot discovers a place he does not know, he recruits a place cell in the entorhinal cortex of the third module. The first module (210) makes it possible to associate this place cell with actions, that is to say to associate allothetic information with grid cells. The process includes as input environmental sensory information, plus particularly visual information (410) from the panoramic camera of the robot. A treatment of received images makes it possible to extract bitters, that is to say, remarkable visual references, easily recognizable in the environment. Various techniques are known from the prior art for extracting this type of information from an image. The general principle is to apply one or more filters on the image and deduce remarkable points by their contrast, brightness or hue and extract the image portions around each point of interest. These visual cues are positioned in the environment of the robot, for example by determining their angular position in the field of view of the robot. By way of non-limiting example, 5 bitterers at defined azimuthal positions uniquely define the planar position of the robot. The azimuthal positions (41 1) of said bitters are determined by image analysis or with the aid of a determined movement of the panoramic camera. Thus all the bitters recruited in the first module constitute a digital compass allowing the robot to determine its position in the environment. The visual and azimuthal information of the bitters is recorded in the robot's memory by recruiting neurons in a PrPh neuron field (420) neuromimetic of the perirhinal and parahippocampal cortex. A place is completely characterized by its image in this matrix (420). In the learning phase, the first module (210) learns to predict the configuration of current grid cells corresponding to the new location in which it is located, as they are identified from purely idiothetic information, using the recruited neurons. in the PrPh (420). Thus, at the end of this training, a grid cell matrix is associated with each location visually recognized by the activity of PrPh neurons (420). This learning is for example realized by means of an algorithm minimizing the quadratic differences and according to the learning rule Widrow and Hoff, described in ". Gaussier, P., Banquet, JP, Sargolini, F., Giovannangeli, C, Save , E., and Poucet, B. (2007). A model ofgrid cells involving extra hippocampal path integration, and the hippocampal loop. Journal of Integrative Neuroscience, 6 (3): 447-476. 56, 61, 62, 90, 91, 92, 93, 114 ". For each grid an independent learning is done.

Returning to FIG. 2, both in the learning phase and in the use phase of the robot, the fourth module (240) performs the fusion of the grid cell batteries. (202) issued from the second module (220) and identified from purely idiothetic information, with the grid cells (201) predicted from the PrPh configuration of the first module, that is to say allothetic information . The fourth module (240) inputs a grid cell battery (202) constructed solely with idiothetic information and a grid cell battery (201) constructed solely with allothetic information. Thus, this fourth module (240) carries out the gate grid fusion, the two batteries (201, 202) of cells to obtain a more precise and robust robot location. The system calculates, for each pair of grids (201, 202), the distance between the position given by the allothetic information and that given by the idiothetic information. This distance is sent into a two-dimensional dynamic neuron field governed by the Amari equation, as described in "Amari. Dynamics of atern formation in lateral-inhibition type neural fields. Biological cybernetics, 27 (2): 77-87, 1977. Said module (240) realizes a temporal integration of this distance and a competition mechanism makes it possible to retain the best estimate of the distance. The activity of the fused grid cell battery (203) is learned in a last group of neurons modeling the entorhinal cortex of the third module (230). Each neuron of the entorhinal cortex represents a cell of place. Each neuron of the entorhinal cortex reacts, by its activity, to the configuration of grid cells (203) that was present at the time of its recruitment.

In use, when the robot travels to known places, there is no recruitment of new neurons either in the entorhinal cortex, or in the PrPh. There is also no recruitment of new landmarks. However, since the environment is dynamic, the system adapts its learning in order to maintain the coherence of its behavior. For this, he adapts the synaptic weights of the connections between the neurons of the different groups. The first module (210) continuously receives grid cell configurations (203) resulting from the merger performed by the fourth (240) module. It updates the synaptic weights between PrPh and predictions of grid cells according to the quadratic difference minimization algorithm. This adaptation makes it possible to correct incorrect responses induced by changes in the environment in order to preserve coherent grid cell predictions. If the grid cell configurations (203) drift between two In order to maintain a consistent prediction, the third module continuously adapts the synaptic weights between the grid cell configurations (203) from the fourth module (240) and the entorhinal cortex of the third module (230).

The exemplary embodiment described above shows the robot's training vis-à-vis the visual allothetic information. However, according to an exemplary embodiment, the allothetic sensory recognition of the places is likely to implement other complementary sensory acquisitions, at least to associate levels of environmental parameters to place cells. Thus, the robot object of the invention is able to contextualize the level of its sensory perceptions according to the place visited, all this by learning and autonomously. For example, the robot object of the invention will consider as "normal" a higher ozone in a room with many printers. Not because this room contains printers but because during these successive passages in this room he finds this rate. Conversely, he will consider as "abnormal" a lower ozone level in another room where this gas is not usually detected.

Thus, according to an exemplary embodiment where the vision is used as a sensor for the recruitment of allothetic indices, the measurements from the other sensors are classified by categories or sensor combinations and the measured level configuration for each category is associated with a battery ( 203) of grid cells from the fourth (240) module. By an inverse model, the current location, represented by the activity of the place cells in the third module (230), and the time, predict the readings of the various environmental sensors. Thus, the robot builds a predictive model of the readings of the different environmental sensors expected in the current location at the current time. The difference between the prediction, given by the sensory context, of the expected levels of the measurements delivered by the different environmental sensors and the actual level measured by these sensors, determines the abnormal situations for which the robot must intervene. Advantageously, when an environmental context anomaly is detected, the robot object of the invention transmits information to its base via the wireless network to which it is connected. This information is advantageously supplemented by position information of the robot. For example, the position of the robot is known by the path taken by said robot to its position current from a known fixed bridge and identified, for example, its charging base, or any other remarkable place recognized by a combination of bitters. Thus, the position of the robot and the location of the alarm can be located on a map without any mapping being recorded in the robot and without specific positioning means, such as beacons, being installed in the environment. .

Returning to FIG. 1, in one embodiment of the invention, the mobile robot (100) purifies the air after detecting the presence of toxic products or at least an atmosphere of poor quality. that is, after establishing the contextualized diagnosis of the ambient atmosphere. Indeed, if the robot detects that one or more of the measured parameters exceed predefined tolerance thresholds, these thresholds being contextualized according to the place or time, then, its computer (150) automatically triggers the setting. operating the reactor (180) to purify the air. According to an exemplary implementation, after the air has been purified, the on-board computer (150) controls the perfume diffuser (170) to perfume the enclosed environment. According to an exemplary embodiment, said perfume diffuser (170) diffuses a dry perfume which makes the atmosphere more pleasant. The air pollution control module (180) and the fragrance diffuser (170) being mounted on the mobile robot (100), the atmosphere of the entire enclosed environment is sanitized. Indeed, the regulation of the atmosphere is a homeostasis operation that is performed in parallel with the diagnostic operation. The mobile robot of the invention is thus able to regulate the atmosphere of the enclosed environment to maintain it in an optimal state.

According to another embodiment, in the case where an environmental signal is detected (smoke, pollutant, odor, etc.), the robot measures at each instant the detected value in order to evaluate the direction of the variation gradient and thus to go back to the source and therefore be as efficient as possible: the robot processes the air continuously by moving in the direction of the source of pollution or towards the center of the residual cloud (if the source is not reachable). Several algorithms of rise of the gradient are usable for this purpose and whose implementation is made possible thanks to the autonomy of the robot in its learning of places:

- move in a straight line as the gradient increases and change direction randomly in the opposite case,

- use the values on three non-aligned positions to estimate the direction of the gradient. This displacement and gradient information is stored and associated with the locations learned by the robot so as to be more effective the next times if the treatment is to be chronic. According to an exemplary embodiment, the method for learning and recognizing a place is triggered by the detection of a significant variation of one of the environmental parameters enabling the robot to learn to locate the areas to be treated.

According to another exemplary embodiment, the ability of raising the gradient is combined with random exploration to allow the robot to be operational in an environment that has never been explored.

In one embodiment of the invention, the human / robot interface comprises an avatar embodying the robot. Such an avatar can facilitate and enrich the interactions between humans and the mobile robot. According to an exemplary embodiment, this avatar is, for example, in the form of a non-humanoid face embodying the robot and displayed on the screen (150). This face is then able to reproduce expressions (sadness, joy, anger, boredom, surprise, fear, etc.), by means of different postures. The expression of this face thus reflects the internal state of the robot (for example: signaling a sensor failure, discovering a new place, empty battery) and punctuates the interaction with the human (for example: smiling at the initiation of the interaction, nodding at the receipt of an order).

Not only does the avatar facilitate man-robot interactions, but it also informs about the current task or the task to be performed. It allows the operator to have a feedback (or feedback, in English terms) during the learning phases. Indeed, by his emotion, he can reflect the success or failure of learning the task.

The description above and the exemplary embodiments show that the invention achieves the desired objectives, in particular, thanks to a neuromimetic navigation system, the robot that is the subject of the invention is autonomous and able to monitor a changing environment according to contextualized parameters. . Thus, the robot object of the invention is capable of creating sensory contexts crossing the information of multiple sensors with its location in the environment and the time at which it visits the area. The robot object of the invention thus learns the regularities of the environment and is capable of detecting and treating abnormal situations.

Claims

Mobile robot (100) comprising displacement means (1 10) and orientation means and sensory means comprising:

at. sensory means capable of delivering proprioceptive information from means (1 10) of displacement and orientation;

b. sensory means (130, 141, 142, 143, 144, 145, 146, 147) adapted to deliver a so-called environmental information perceived in the environment of the robot;

characterized in that it comprises computer means comprising: c. a neuromimetic structure (210, 220, 230) comprising a battery of idiothetic gate cells (202) whose neurons are activated by proprioceptive information and a battery of allothetic gate cells (201);

Robot according to claim 1, comprising:

e. computer means (240) for merging the information of the allothetic gate cell bank (201) and the idiothetic gate cell bank (202).

Robot according to claim 2, comprising:

f. a neuromimetic structure, called PrPh (420), whose neurons are activated by the combination of proprioceptive information (41 1) and environmental information (410).

Robot according to claim 1, wherein the proprioceptive sensory means comprise one or more of:

an odometer;

an orientation sensor;

a speed sensor.

The robot of claim 1, wherein the environmental sensory means comprises one or more of:

a vision sensor (130);

a thermometer (146);

- a dust detector (142);

a humidity detector (143);

a toxic product detector (145);

a gas detector (141);

a smoke detector (147);

- a microphone.

The robot of claim 5, wherein the vision sensor (130) is a panoramic camera and includes proprioceptive means for measuring the viewing orientation of said camera.

7. Robot according to claim 6, comprising means, said digital compass, capable of determining the position of the robot relative to an allothetic mark.

8. Robot according to claim 1, comprising a plurality of proximity sensors (160) capable of preventing a collision.

9. Robot according to claim 5, comprising means (170, 180), said effectors, able to modify the environment of the robot.

The robot of claim 9 wherein the effector means comprises:

a fragrance diffuser (170); or

a chemical or biological depollution reactor (180); or - an air humidifier.

11. Robot according to claim 5, comprising alarm means (191, 192).

12. Method for navigating in a closed environment by means of a robot according to claim 3, characterized in that it comprises the learning steps consisting in:

i. recruit a place cell neuron into the entorhinal cortex; ii. recruit environmental clues using the environmental sensor and merge these clues with their position in the robot environment by recruiting neurons into PrPh (420);

iii. associating with this configuration of the PrPh the configuration of the grid cells (203);

The method of claim 12, comprising the autonomous navigation steps of:

iv. merging grid gate idiothetic grid cells derived from information from proprioceptive sensory means and grid cells derived from predictions of PrPh;

vi. to determine the cell of the most active place of the entorhinal cortex according to the cells of gilles resulting from the fusion;

vii. to deduce the position of the robot and the sequence of actions.

The method of claim 13 including a navigation step in which the robot enters an unexplored location and which comprises steps i) to iii) so as to integrate that location into its memory.

A method according to claim 13, employing a robot according to claim 5, wherein the environmental indicia recruited in step ii) are visual bitters and which comprises the steps of:

viii. recruiting environmental indices using other environmental sensors;

ix to be associated with the configuration of the environmental indics recruited in step viii) the configuration of the grid cells (203).

The method of claim 15, comprising a step of: x. categorizing the environmental indices recruited in step viii) and that the association of grid cells (203) of step ix) is performed by category.

17. The method of claim 15, employing a robot according to claim 10 or claim 11, comprising the steps of:

xi. if the information from a sensory environment means deviates significantly from the level known in this place;

xii. act on the environment by means of an effector or trigger an alarm.

The method of claim 17, comprising a step of xiii. moving the robot into the environment following the gradient of the information detected in step xi) to go back to the source of the emission.