Safe robot execution in model-based reinforcement learning

Task learning in robotics requires repeatedly executing the same actions in different states to learn the model of the task. However, in real-world domains, there are usually sequences of actions that, if executed, may produce unrecoverable errors (e. g. breaking an object). Robots should avoid repeating such errors when learning, and thus explore the state space in a more intelligent way. This requires identifying dangerous action effects to avoid including such actions in the generated plans, while at the same time enforcing that the learned models are complete enough for the planner not to fall into dead-ends.

Authors

• David Martínez 0004
• Guillem Alenyà
• Carme Torras

Keywords

• Safety
• Planning
• Context
• Learning (artificial intelligence)
• Space exploration
• Manipulators
• Model-based Reinforcement Learning
• Risky Activities
• Alternative Plans
• Success Ratio
• Robot Learning
• Decision-making
• Learning Algorithms
• Shortest Path
• Markov Decision Process
• Reinforcement Learning Algorithm
• Acceptable Risk
• Interactive Teaching
• Real Robot
• State-action Pair