2018/03/21 by Wissam Bejjani, Bejjani, Wissam, Rafael Papallas +5
Computer Science · Engineering · #AI-based Problem Solving and Planning #FOS: Computer and information sciences #Formal Methods in Verification #Reinforcement Learning in Robotics #Robot Manipulation and Learning #Robotic Path Planning Algorithms #Robotics (cs.RO)
paper · pdf · doi:10.48550/arxiv.1803.08100
openalex publication_date 2018/03/21 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Manipulation in clutter requires solving complex sequential decision making\nproblems in an environment rich with physical interactions. The transfer of\nmotion planning solutions from simulation to the real world, in open-loop,\nsuffers from the inherent uncertainty in modelling real world physics. We\npropose interleaving planning and execution in real-time, in a closed-loop\nsetting, using a Receding Horizon Planner (RHP) for pushing manipulation in\nclutter. In this context, we address the problem of finding a suitable value\nfunction based heuristic for efficient planning, and for estimating the\ncost-to-go from the horizon to the goal. We estimate such a value function\nfirst by using plans generated by an existing sampling-based planner. Then, we\nfurther optimize the value function through reinforcement learning. We evaluate\nour approach and compare it to state-of-the-art planning techniques for\nmanipulation in clutter. We conduct experiments in simulation with artificially\ninjected uncertainty on the physics parameters, as well as in real world tasks\nof manipulation in clutter. We show that this approach enables the robot to\nreact to the uncertain dynamics of the real world effectively.\n