2021/05/20 by Alex Mitrevski, Mitrevski, Alex, Paul G. Plöger +3
Computer Science · Engineering · #Explainable Artificial Intelligence (XAI) #FOS: Computer and information sciences #Reinforcement Learning in Robotics #Robot Manipulation and Learning #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.2105.09599
openalex publication_date 2021/05/20 · openalex created_date 2022/07/25 · openalex updated_date 2026/07/28
When faced with an execution failure, an intelligent robot should be able to\nidentify the likely reasons for the failure and adapt its execution policy\naccordingly. This paper addresses the question of how to utilise knowledge\nabout the execution process, expressed in terms of learned constraints, in\norder to direct the diagnosis and experience acquisition process. In\nparticular, we present two methods for creating a synergy between failure\ndiagnosis and execution model learning. We first propose a method for\ndiagnosing execution failures of parameterised action execution models, which\nsearches for action parameters that violate a learned precondition model. We\nthen develop a strategy that uses the results of the diagnosis process for\ngenerating synthetic data that are more likely to lead to successful execution,\nthereby increasing the set of available experiences to learn from. The\ndiagnosis and experience correction methods are evaluated for the problem of\nhandle grasping, such that we experimentally demonstrate the effectiveness of\nthe diagnosis algorithm and show that corrected failed experiences can\ncontribute towards improving the execution success of a robot.\n