| 研究実績の概要 |
In FY2017 we investigated abstraction methods to simplify the motion planning problem in robotics. We proposed a new method of nesting robots in each other, which translates to nesting a sequence of quotient spaces in the configuration space of the robot. Each quotient space abstracts away some information from the robot, for example a robot described by position and orientation would have a quotient space described solely by its position.
After constructing such a quotient space decomposition for a robot, we developed a new algorithm called the Quotient Space RoadMap Planner (QMP) which can exploit such a decomposition. QMP starts at the lowest quotient space level, and constructs a graph to connect start and goal configuration. Once such a graph has been found, QMP constructs a new graph in the next bigger quotient space, but constrained by the edges of the underlying graph. In that way we can sample the configuration space in a more efficient way, which translates to a faster runtime of our algorithm compared to state-of-the-art planning algorithms. We have proven that QMP is probabilistically complete, this means that for any planning problem, the algorithm will find a path if one exists as the time goes to infinity.
We have demonstrated experimentally that QMP is faster than RRT, PRM and EST (three state-of-the-art planning algorithms) on four different scenarios. We showed the applicability to a rigid body free floating in space, to an articulated body free floating in space, to a fixed-base manipulators.
|
