Download presentation
Presentation is loading. Please wait.
Published byEleanor Robbins Modified over 8 years ago
1
Chapter 12. Some Requirements for Human-Like Robots: Why the Recent Over-Emphasis on Embodiment Has Held Up Progress Aaron Sloman School of Computer Science, University of Birmingham Course: Robots Learning from Humans DooSan Baek Vehicle Intelligence Laboratory School of Electrical Engineering Seoul National University http://vi.snu.ac.kr
2
Contents Introduction The Seduction of Embodiment Fallacies in Nouvelle AI Limitations of Symbolic AI Meta-semantic and Exosomatic Ontologies Morphology and Development 2
3
Introduction There are important limitations in current ideas about embodiment and dynamical systems Only considers sense-think-act cycle Rejects symbolic AI Their failure is mainly because.. Problems were not understood Not that the wrong mechanisms were adopted 3
4
The Seduction of Embodiment Symbolic AI Symbol grounding Kant’s empiricism Learning of robots to discovery of statistical patterns relating sensory and motor signals Nouvelle AI (Brook) Emphasis embodiment and sensory-motor interactions with the environment Dispense with symbolic representations (But not fully rejected) Use morphology to reduce the SW sophistication 4
5
The Seduction of Embodiment Morphology Can be Important Simplify problems of grasping and manipulation of 3D objects Recent work on embodied robots has been impressive BigDog, Asimo While still failing Plan future actions Reason about unobserved events Represent mental states of others 5
6
Fallacies in Nouvelle AI The Failure to Identify Deliberative Requirements Ex) Elephant The immediate environment may be a good source But it is not suitable for complex decision making Elephant may not play chess, solve algebra problem Simple but sophisticated framework is needed “Fully deliberative” 6
7
Fallacies in Nouvelle AI Fully deliberative Construct several representations of branching futures Represent and compare their relative strengths and weakness Select one as a plan and use it to generate behaviors While being prepared to learn from mistaken decisions of that sort 7
8
Limitations of Symbolic AI The early symbolic AI Did not take account sufficient uncertainty and error Did not consider actions that causing unreliable results Slippage, wear and tear 8
9
Meta-semantic and Exosomatic Ontologies Meta-semantic Robot should understand other intelligent agent’s actions Robot should infer other’s mental states Exosomatic Ontology For sophisticated animals Allows generalizations For perceiving a grasping One hand -> other hand, two hand, other agents 9
10
Morphology and Development Morphology and perception Morphology is less relative with the ability of environment perception Thalidomide babies, SLAM Human can overcome deformity, illness 10
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.