End-to-end Deep Learning-Based Framework for Path Planning and Collision Checking: Bin Picking Application

Real-time and efficient path planning is critical for all robotic systems. Inparticular, it is of greater importance for industrial robots since the overallplanning and execution time directly impact the cycle time and automationeconomics in production lines. While the problem may not be complex in staticenvironments, classical approaches are inefficient in high-dimensionalenvironments in terms of planning time and optimality. Collision checking posesanother challenge in obtaining a real-time solution for path planning incomplex environments. To address these issues, we propose an end-to-endlearning-based framework viz., Path Planning and Collision checking Network(PPCNet). The PPCNet generates the path by computing waypoints sequentiallyusing two networks: the first network generates a waypoint, and the second onedetermines whether the waypoint is on a collision-free segment of the path. Theend-to-end training process is based on imitation learning that uses dataaggregation from the experience of an expert planner to train the two networks,simultaneously. We utilize two approaches for training a network thatefficiently approximates the exact geometrical collision checking function.Finally, the PPCNet is evaluated in two different simulation environments and apractical implementation on a robotic arm for a bin-picking application.Compared to the state-of-the-art path planning methods, our results showsignificant improvement in performance by greatly reducing the planning timewith comparable success rates and path lengths.