Disassembly sequencing using a motion planning approach

We propose a new approach for generating disassembly sequences. Our motion planning based approach treats the parts in the assembly as robots and operates in the composite configuration space of the parts' individual configuration spaces. Randomized techniques inspired by recent motion planning methods are used to sample configurations in this space. Since typical assemblies consist of many parts, the corresponding composite C-spaces have high dimensionality. Also, since many important configurations for the disassembly sequence will involve closely packed parts, the disassembly problem suffers from the so-called narrow passage problem. Since random sampling in this situation would not be effective, we bias the sampling by computing potential movement directions based on the geometric characteristics of configurations known to be reachable from the assembled configuration (the start). For example, we select potential directions which are perpendicular to part faces. Thus, we construct a disassembly tree which is rooted at the starting assembled configuration. Our experimental results with several non-trivial puzzle-like assemblies (12-72 dof) show the potential of this approach.