by F Stulp and M Beetz
Abstract:
As agent systems are solving more and more complex tasks in increasingly challenging domains, the systems themselves are becoming more complex too, often compromising their adaptivity and robustness. A promising approach to solve this problem is to provide agents with reflective capabilities. Agents that can reflect on the effects and expected performance of their actions, are more aware and knowledgeable of their capabilities and shortcomings. In this paper, we introduce a computational model for what we call \textbackslashemphaction awareness. To achieve this awareness, agents learn predictive action models from observed experience. This knowledge is then used to optimize, transform and coordinate plans. We apply this computational model to a number of typical scenarios from robotic soccer. Various experiments on real robots demonstrate that action awareness enables the robots to improve the performance of their plans substantially.
Reference:
Action Awareness – Enabling Agents to Optimize, Transform, and Coordinate Plans (F Stulp and M Beetz), In Proceedings of the Fifth International Joint Conference on Autonomous Agents and Multiagent Systems (AAMAS), 2006.
Bibtex Entry:
@inproceedings{stulp_action_2006,
author = {F Stulp and M Beetz},
title = {Action Awareness – Enabling Agents to Optimize, Transform, and Coordinate
Plans},
booktitle = {Proceedings of the Fifth International Joint Conference on Autonomous
Agents and Multiagent Systems ({AAMAS)}},
year = {2006},
abstract = {As agent systems are solving more and more complex tasks in increasingly
challenging domains, the systems themselves are becoming more complex
too, often compromising their adaptivity and robustness. A promising
approach to solve this problem is to provide agents with reflective
capabilities. Agents that can reflect on the effects and expected
performance of their actions, are more aware and knowledgeable of
their capabilities and shortcomings. In this paper, we introduce
a computational model for what we call {\textbackslash}emphaction
awareness. To achieve this awareness, agents learn predictive action
models from observed experience. This knowledge is then used to optimize,
transform and coordinate plans. We apply this computational model
to a number of typical scenarios from robotic soccer. Various experiments
on real robots demonstrate that action awareness enables the robots
to improve the performance of their plans substantially.},
}