Panoramic attention for humanoid robots

• US 8,406,925 B2
• Filed: 06/18/2010
• Issued: 03/26/2013
• Est. Priority Date: 07/01/2009
• Status: Active Grant

• Current Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Original Assignee: Honda Motor Co., Ltd. (Honda Motor Company)
• Inventors: Sarvadevabhatla, Ravi Kiran, Ng-Thow-Hing, Victor
• Primary Examiner(s): Cheung, Mary
• Assistant Examiner(s): LANG, MICHAEL DEAN
• Application Number:
  US12/819,032
• Publication Number:
  US 20110004341A1
• Time in Patent Office: 1,012 Days
• Field of Search: 700/250, 700/259
• US Class Current: 700/259
• CPC Class Codes: B25J 9/1602 characterised by the control system, structure, architecture
• CPC Class Codes: G05B 2219/39387 Reflex control, follow movement, track face, work, hand, visual servoing
• CPC Class Codes: G05B 2219/40004 Window function, only a specific region is analyzed
• CPC Class Codes: G05B 2219/40536 Signal processing for sensors

Abstract

A robot using less storage and computational resources to embody panoramic attention. The robot includes a panoramic attention module with multiple levels that are hierarchically structured to process different levels of information. The top-level of the panoramic attention module receives information about entities detected from the environment of the robot and maps the entities to a panoramic map maintained by the robot. By mapping and storing high-level entity information instead of low-level sensory information in the panoramic map, the amount of storage and computation resources for panoramic attention can be reduced significantly. Further, the mapping and storing of high-level entity information in the panoramic map also facilitates consistent and logical processing of different conceptual levels of information.

Claims

1. A method of operating a robot, comprising:

extracting features from a sensor signal responsive to receiving the sensor signal from a sensor;

determining one or more portions of the sensor signal likely to include one or more entities based on the extracted features;

detecting one or more entities in the sensor signal by processing the one or more portions of the sensor signal;

generating entity information indicating characteristics of each of the detected entities;

determining a location of each of the detected entities relative to the robot in a panoramic map at least partially surrounding the robot based on the entity information, the location of each of the detected entities represented by a coordinate in the panoramic map; and

controlling the robot to perform an action based on the detected entities and the location determined for each of the detected entities.

2. The method of claim 1, wherein the panoramic map identifies locations using a spherical coordinate system.

3. The method of claim 1, further comprising storing a trace of a detected entity representing changes in a location of the detected entity with progress of time.

4. The method of claim 1, wherein each entity is associated with an activation signal representing confidence or probability that the entity information and the location of a previously detected entity remains valid, an activation signal of a previously detected entity decreased with progress of time while a sensory field of the sensor does not cover the previously detected entity.

5. The method of claim 4, wherein a first activation signal for a first type of entity is decreased at a first rate and a second activation signal for a second type of entity is decreased at a second rate.

6. The method of claim 4, wherein controlling the robot comprises causing the robot to move to include a location of an entity previously detected in a sensory field of the sensor responsive to detecting an event associated with the previously detected entity.

7. The method of claim 1, wherein the sensor is an image capturing device and the sensor signal comprises a stream of images.

8. The method of claim 7, wherein the extracted features comprises one or more of light intensity, colors, motions, and orientations in the stream of images.

9. The method of claim 7, further comprising:

generating feature maps representing relationships between locations and features in an image of the stream of images; and

generating saliency maps by combining the feature maps, the one or more portions of the sensor signal determined based on the saliency maps.

10. The method of claim 7, further comprising selecting a facing location of the robot based on previous locations the image capturing device faced in response to the robot being placed in an idle mode, a facing location previously selected having a lower probability of being selected.

11. The method of claim 7, further comprising rectifying the image captured by the image capturing device to eliminate spherical distortions in the captured image.

12. The method of claim 7, further comprising:

extracting features of audio signals received at microphones;

determining locations of the audio signal;

extracting features of the audio signals; and

storing the audio features in association with the locations of the audio signals.

13. The method of claim 1, further comprising

planning the action to achieve a goal;

and

modifying, based on the goal, at least one of (i) the features extracted from the sensor signal, (ii) one or more portions of the sensor signal selected and (iii) types of entities detected.

14. A robot comprising:

an attention module comprising,

a low-level sub-module configured to extract features from a sensor signal responsive to receiving the sensor signal from a sensor;

a mid-level sub-module configured to;

determine one or more portions of the sensor signal likely to include one or more entities based on the extracted features,

detect one or more entities in the sensor signal by processing the one or more portions of the sensor signal; and

generate entity information indicating characteristics of each of the one or more detected entities; and

a top-level sub-module configured to determine a location of each of the one or more detected entities relative to the robot; and

map the detected entities to locations in a panoramic map at least partially surrounding the robot based on the entity information, the location of each of the detected entities represented by a coordinate in the panoramic map; and

a behavior module configured to control the robot to perform an action based on the detected entities and the determined location for each of the detected entities.

15. The robot of claim 14, wherein the attention module further comprises:

an entity storage configured to store, for each of the detected entities, the entity information, the location, and an activation signal representing confidence or probability that the entity information and the location of a previously detected entity remains valid; and

an entity manager configured to decrease the activation signal of the previously detected entity with progress of time while a sensory field of the sensor does not cover the previously detected entity.

16. The robot of claim 15, wherein the entity storage is further configured to store a trace of a detected entity, the trace representing changes in a location of the detected entity with the progress of time.

17. The robot of claim 14, wherein the behavior module is configured to cause the robot to move to include a location of an entity previously detected in a sensory field of the sensor responsive to detecting an event associated with the previously detected entity.

18. The robot of claim 14, wherein the sensor is an image capturing device and the sensor signal comprises a stream of images.

19. The robot of claim 18, wherein the extracted features comprises one or more of light intensity, colors, motions, and orientations in the stream of images.

20. The robot of claim 18, wherein the low-level sub-module is further configured to generate feature maps and saliency maps, the feature maps representing relationships between locations and features in an image of the stream of images, the saliency maps generated by combining the feature maps, and the mid-level sub-module is configured to determine the one or more portions of the sensor signal based on the saliency maps.

21. The robot of claim 18, wherein the behavior module is further configured to select a facing location of the robot based on previous locations the image capturing device faced in response to the robot being placed in an idle mode, a previously selected facing location having a lower probability of being selected in current selection.

22. The robot of claim 18, further comprising a pre-processor for rectifying the image captured by the image capturing device to eliminate spherical distortions in the captured image, the rectified image provided to the low-level sub-module for detecting the features.

23. The robot of claim 14, wherein the behavior module is configured to plan the action to achieve a goal, and modify, based on the goal, at least one of:

(i) the features extracted from the sensor signal, (ii) one or more portions of the sensor signal selected and (iii) types of entities detected.

24. A computer readable storage medium configured to operate a robot, the computer readable storage medium structured to store instructions, when executed, cause a processor in the robot to:

extract features from a sensor signal responsive to receiving the sensor signal from a sensor;

determine one or more portions of the sensor signal likely to include one or more entities based on the extracted features;

detect one or more entities in the sensor signal by processing the one or more portions of the sensor signal;

generate entity information for each of the detected entities, the entity information indicating characteristics of each of the one or more detected entities;

determine a location of each of the detected entities relative to the robot in a panoramic map at least partially surrounding the robot based on the entity information, the location of each of the detected entities represented by a coordinate in the panoramic map; and

control the robot to perform an action based on the detected entities and the location determined for each of the detected entities.