Towards High-performance Robot Plans with Grounded Action Models: Integrating Learning Mechanisms into Robot Control Languages

Alexandra Kirsch (1, 2)

1 Intelligent Autonomous Systems
2 Human-Computer Interaction and Artificial Intelligence

For planning in the domain of autonomous robots, abstraction of state and actions is indispensable. This abstraction however comes at the cost of suboptimal execution, as relevant information is ignored. A solution is to maintain abstractions for planning, but to fill in precise information on the level of execution. To do so, the control program needs models of its own behavior, which could be learned by the robot automatically. In my dissertation I develop a robot control and plan language, which provides mechanisms for the representation of state variables, goals and actions, and integrates learning into the language.

Domaines: Intelligence artificielle [cs.AI]; Apprentissage [cs.LG]; Robotique [cs.RO]

Alexandra Kirsch. Towards High-performance Robot Plans with Grounded Action Models: Integrating Learning Mechanisms into Robot Control Languages. International Conference on Automated Planning & Scheduling (ICAPS), Jun 2005, Monterey, CA, United States. pp.47-49. ⟨hal-01329024⟩