Adaptive Path Planning for Reaching an Uncertain Set of Targets in a Fruit Tree

When inspecting a set of target points (such as fruit) in a fruit tree, it is not realistic to assume that a perfectly-accurate set of such targets is available beforehand: compared to earlier inspections, new fruit may have appeared, and some fruit may be missing. In previous work, we introduced a shell-based method for rapid global path planning to position the end of a robotic arm close to numerous target points in an environment for inspection. In this work, we extend our method to handle partial knowledge of the set of targets, enabling the robot to adapt its planned path to newly-discovered or missing fruit on-the-fty. In the most extreme case, the robot starts with no knowledge of the workspace at all, except for what is immediately visible on startup. We show that, when not all targets are given beforehand, there is only a negligible increase in the length of the path traversed by the robot, the robot still effectively discovers a-priori unknown targets, at an overall about 35% increase in the CPU running time of the algorithm compared to the full-knowledge scenario.