A dynamic planner for object assembly tasks based on learning the spatial relationships of its parts from a single demonstration

In this paper, we propose a general system for enabling robots to generate assembly plans for assisting people during assembly tasks. Such a plan is derived from a 3D occupancy grid that the system generates while observing a person performing an assembly task. Our proposed system uses the acquired 3D occupancy grid and a graph search to generate an assembly plan. This plan is used to guide users during assembly tasks to create a similar object. If the user deviates from the suggested plan, our system automatically validates whether the new state is solvable or not and reacts accordingly. Forward assembly planning is an NP-hard problem, but we introduce pruning methods for the search tree that make the approach practical.