1 / 44

A Cognitive Map for an Artificial Agent

A Cognitive Map for an Artificial Agent. Unmesh Kurup RPI kurupu@rpi.edu B. Chandrasekaran The Ohio State University chandra@cse.ohio-state.edu. Overview. Cognitive map Features Goals biSoar architecture Cognitive Map in biSoar Examples. Cognitive Map. Cognitive Map.

tilden
Télécharger la présentation

A Cognitive Map for an Artificial Agent

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. A Cognitive Map for an Artificial Agent Unmesh Kurup RPI kurupu@rpi.edu B. Chandrasekaran The Ohio State University chandra@cse.ohio-state.edu

  2. Overview • Cognitive map • Features • Goals • biSoar architecture • Cognitive Map in biSoar • Examples

  3. Cognitive Map

  4. Cognitive Map • Representation of large-scale space

  5. Cognitive Map • Representation of large-scale space • Layout of a city

  6. Cognitive Map • Representation of large-scale space • Layout of a city or a building

  7. Cognitive Map • Representation of large-scale space • Layout of a city or a building • Supports a number of problem solving tasks.

  8. Cognitive Map • Representation of large-scale space • Layout of a city or a building • Supports a number of problem solving tasks. • Route-finding: How can I get to the Radisson from here? • Exit the hotel, Take a left, Right at 23rd St, Right on Clark.

  9. Cognitive Map • Representation of large-scale space • Layout of a city or a building • Supports a number of problem solving tasks. • Geo Recall: Is your hotel to the west of this hotel?

  10. Cognitive Map • Representation of large-scale space • Layout of a city or a building • Supports a number of problem solving tasks. • Finding shortcuts: Is there a shorter way to my hotel? • Possible: Take a right on 20th st.

  11. Features of the Cognitive Map

  12. Features of the Cognitive Map • Non-holistic

  13. Features of the Cognitive Map • Non-holistic If you take a left on Crystal Dr, you will get to the intersection of Crystal Dr and 23rd St If you take a right on 23rd St, you will get to the intersection of 23rd St and Clark.

  14. Features of the Cognitive Map • Non-holistic • manageable

  15. Features of the Cognitive Map • Non-holistic • Manageable

  16. Features of the Cognitive Map • Non-holistic • Manageable If you take a left on Crystal Dr, you will get to the intersection of Crystal Dr and 23rd St If you take a right on 23rd St, you will get to the intersection of 23rd St and Clark. vs If you take a right on Clark you will get to your hotel

  17. Features of the Cognitive Map • Non-holistic • Manageable, updateable,

  18. Features of the Cognitive Map • Non-holistic • Manageable, updateable, If you take a left on Crystal Dr, you will get to the intersection of Crystal Dr and 23rd St If you take a right on 20th St, you will get to the intersection of 20th St and Clark. If you take a right on 23rd St, you will get to the intersection of 23rd St and Clark. New info? Just add it! If you take a right on Clark you will get to your hotel

  19. Features of the Cognitive Map • Non-holistic • Manageable, updateable, composable

  20. Features of the Cognitive Map • Non-holistic • Manageable, updateable, composable If you take a left on Crystal Dr, you will get to the intersection of Crystal Dr and 23rd St If you take a right on 20th St, you will get to the intersection of 20th St and Clark. If you take a right on Clark you will get to your hotel

  21. Features of the Cognitive Map • Non-holistic • Manageable, updateable, composable • Has both symbolic and metric aspects

  22. Features of the Cognitive Map • Non-holistic • Manageable, updateable, composable • Has both symbolic and metric aspects If you take a left on Crystal Dr, you will get to the intersection of Crystal Dr and 23rd St

  23. Goals • Capture features • A non–holistic representation with both symbolic and metric aspects

  24. Goals • Capture features • A non–holistic representation with both symbolic and metric aspects • Cognitive architecture approach

  25. biSoar A Bimodal Cognitive Architecture

  26. biSoar • Soar + Diagrammatic Representation System (DRS)

  27. DRS - Diagrammatic Representation System (Chandra et. al. 2004) • Diagrams consist of three types of objects – Points, Curves & Regions.

  28. DRS - Diagrammatic Representation System (Chandra et. al. 2004) • Perceptual routines allow extraction of relationships between objects in the diagram. • Ex: LeftOf, RightOf etc • Action routines allow the diagram to be modified • AddPoint, AddCurve etc

  29. Working Memory Symbolic component: Block (A), Block (B), Block (C), On (A,B), On (B,C) Selected Operator: None Diagrammatic component A B C biSoar Soar World Working Memory: Block (A), Block (B), Block (C), On (A,B), On (B,C) Selected Operator: None biSoar Soar WM DRS

  30. LTM and Learning in biSoar • No change to LHS of LTM rules in biSoar • RHS can extract information from or modify diagrammatic component as well. • If a and b are clear and goal is on(a,b) then translate(a on b) • Chunking in the bimodal case is straightforward.

  31. biSoar • Soar + Diagrammatic Representation System (DRS) • biSoar does not do • Any sort of image processing • Object recognition • Assumes • a diagrammatic representation (DRS form) of the input is available.

  32. Representing LSS in biSoar If goal is find_next_location and curr location is x and traveling in direction Dx on route Rx, then destination is location y, diagram is DRSx

  33. R4 R5 R1 R3 R2 Representing LSS in biSoar If goal is find_next_location and curr location is A and traveling in direction Dx on route Rx, then destination is location B, diagram is DRSx If goal is find_next_location and curr location is R2R5 and traveling Right on Route R2, then destination is P2, diagram is DRS1

  34. Examples – Route-finding • Given a map, find route from P1 to P2 • Route-finding Strategy • locate the starting & destination locations in the map • make the starting location the current location • Find the routes on which the current location lies • For each route, find the directions of travel • for each route and direction of travel, find the next location • calculate the Euclidean distance between these new locations and the destinations • pick the location that is closest to the destination and make that the current point • repeat 3-8 until destination is reached

  35. Learning while route-finding Example rules learned during wayfinding

  36. R4 R5 R1 R3 R2 Within-task transfer • Task1 – P1 to R3R5

  37. R4 R5 R1 R3 R2 Within-task transfer • Task1 – P1 to R3R5, Task2 – P4 to P2

  38. R4 R5 R1 R3 R2 Within-task transfer • Task1 – P1 to R3R5, Task2 – P4 to P2 • New route-finding task – P4 to R3R5

  39. Between-task transfer • Geographic Recall problem • What’s the spatial relation between R1R3 and R3R5?

  40. Finding short-cuts R4 If goal is find_routes and at R2R5 then there is a route r5 in the up direction. R1 R3 R2

  41. Finding short-cuts R4 If goal is find_routes and at R2R5 then there is a route r5 in the up direction. R1 R3 Find route from P2 to P5 R2

  42. Finding short-cuts R4 If goal is find_routes and at R2R5 then there is a route r5 in the up direction. R1 R3 Find route from P2 to P5 R2 R3 R5 R2

  43. Conclusion • biSoar’s CM (representation of LSS) • Is non-holistic • has metric and non-metric information • Can be used to solve a variety of tasks involving LSS. • Supports learning and transfer of learned information within and between tasks.

  44. Thanks!

More Related