In this paper, we investigate methods for enabling a human operator and an automatic motion planner to cooperatively solve a motion planning query. Our work is motivated by our experience that automatic motion planners sometimes fail due to the difficulty of discovering `critical' configurations of the robot that are often naturally apparent to a human observer. Our goal is to develop techniques by which the automatic planner can utilize (easily generated) user-input, and determine `natural' ways to inform the user of the progress made by the motion planner. We show that simple randomized techniques inspired by probabilistic roadmap methods are quite useful for transforming approximate, user-generated paths into collision-free paths, and describe an iterative transformation method which enables one to transform a solution for an easier version of the problem into a solution for the original problem. We also illustrate that simple visualization techniques can provide meaningful representations of the planner's progress in a 6-dimensional C-space. We illustrate the utility of our methods on difficult problems involving complex 3D CAD Models.
ASJC Scopus subject areas
- Artificial Intelligence