Une approche interactioniste pour des robots adaptatifs auto-motivés

Size: px

Start display at page:

Download "Une approche interactioniste pour des robots adaptatifs auto-motivés"

Marion Allison
6 years ago
Views:

1 Une approche interactioniste pour des robots adaptatifs auto-motivés Olivier GEORGEON Alain Mille Christian Wolf LIRIS 1 /27

2 Outline Challenge Theoretical approach Demonstrations 2 /27

3 Challenge Generate "smart " behavior. Unsolved challenge. ill-defined challenge. It is not about problem-solving! 3 /27

4 Issues with the cognitive-loop paradigm Problem of using subjective concepts Problem of separating perception and action Agent Action Observation Environment The cognitive sandwich Hurley, S. (1998). Consciousness in action. 4 /27

5 Issues with the cognitive-loop paradigm Consider perception as a cognitive construct rather than an input. Agent Movement Effect Sensorimotor scheme Environment Piaget (1937) La construction du réel chez l enfant. 5 /27

6 Issues with the cognitive-loop paradigm Agent Movement De ve tra lop jec me t o nt a ry l Problem of levels of time scales. Effect Environment 6 /27

7 Issues with the cognitive-loop paradigm The unrolled model. ry o t c e aj r t l a nt e m lop e v e D Level rn a Le rn a Le Ac tiva te Ac tiva te Ac tiva te Time 7 /27

8 Issues with the cognitive-loop paradigm The problem of space. It is not only about time and sequences! Agent Effect Movement Environment 8 /27

9 Issues with the cognitive-loop paradigm The spatial unrolled model. ry o t c e aj r t l a nt e m lop e v e D Level rn a Le rn a Le Ac tiva te Ac tiva te Ac tiva te Time 9 /27

Process (POMDP) Interactional Motivation in Artificial Systems: Between

10 Interactional motivation Agent Movement Value Effect Sensorimotor scheme Environment Interactional Motivation Partially Observable Markov Decision Process (POMDP) Interactional Motivation in Artificial Systems: Between Extrinsic and Intrinsic Motivation. Georgeon, Marshall, Gay. Submitted to EpiRob /27

11 Exemple Set of schemes: -Step or bump 5 / -10 -Turn left / turn right -3 -Touch front / left / right -1 Bump: Touch: 11 /27

12 Trace Level Touch Value Touch wall Touch empty Try move forward Left Front Right Move forward Bump Turn left Turn right 12 /27

13 Trace touch front move forward (step 67) touch left turn left move forward (Step 99) 13 /27

14 touch left empty, turn left, move forward, touch front wall. Upper left area Left corner Left empty Right empty Front empty 14 /27

15 Assessment criteria Measurable criteria Objective hedonism Situational categorization Situational disambiguation Behavioral criteria Hedonistic temperance Graceful readaptation Active perception Individuation Designing Environment-Agnostic Agents. Olivier L. Georgeon, Ilias Sakellariou. AAMAS (2012). 15 /27

16 Vision and space 16 /27

17 Stratégie diagonale 17 /27

18 Stratégie tangentielle 18 /27

19 Self model and space Left place s6 s9 Translate spatial memory when s1 s3 s4 s7 s10 Right place Rotate spatial memory when s4 s8 s5 s2 s1 Front place Rotate spatial memory when s3 19 /27

20 Exemple spatial 20 /27

21 Trace Empty phenomenon Wall phenomenon 21 /27

22 Architecture 22 /27

23 From drives to goals Afforded Simulated 23 /27

24 Dynamic and continuous environment 24 /27

25 Example in a robot 25 /27

26 Robot with vision 26 /27

27 Conclusion Novel learning algorithm Developmental approach to cognition Agnostic agents No ontological presupposition on the environment Qualitative evaluation as opposed to performance measures Activity traces Interactional Motivation Between Extrinsic and Intrinsic motivation The Small Loop Problem Still unsolved! 27 /27

Inferring Actions and Observations from Interactions

2013 Annual Conference on Advances in Cognitive Systems: Workshop on Goal Reasoning Inferring Actions and Observations from Interactions Joseph P. Garnier Olivier L. Georgeon Amélie Cordier Université