Efficient, Risk-Encoding Octrees for Path Planning with a Robot Manipulator

Abstract

Recent research in robotics envisions shared human-robot workspaces to combine individual advantages of humans and robots. As part of this vision, robot manipulators must avoid collisions with humans and other a priori unknown obstacles in the shared workspace. State-of-art approaches (e.g. certainty grids, bounding volume hierarchies) test robot poses for collisions under individual limitations (e.g. memory or processing overhead, assumption of global views or noiseless sensors). In contrast, we contribute a sample-based pose test alongside a hierarchical representation of risk over the shared workspace. Our contribution has low memory and processing overhead, and allows for local, outdated, or noisy sensor views. Experiments with real-world data validate this claim and show advantages and limits of our approach over competing variants. We conclude that our pose test and risk representation enhance real-time path planning for robot manipulators in current and future use cases.