Trajectory Deformation


Path deformation is a technique to continuously deform a pre-computed path on-line. In general robotic system, before a robot starts to move, a path planner generates a path for the robot to move from a given initial configuration to a given goal configuration. The robot will then follow this path to accomplish the given task. This strategy performs well in highly-controlled environment such as factory floor. However, when robots move in human environment, the pre-computed path will most likely become invalid due to obstacles that are not present or not known in advance. Although the pre-computed path is no longer valid, computing a new path from scratch is often too costly to do. On the other hand, in general, some part of the pre-computed path are still valid. Therefore, path deformation proposes to reuse the pre-computed path by continuously deforming the path, in accordance to its sensor input, assuming near perfect sensing.

In this project, we propose to extend path deformation to trajectory deformation by taking the time dimension into account when deforming a pre-computed trajectory. This enable us to exploit information about future behavior of the dynamic obstacles. We propose a trajectory deformer called 2-Step Trajectory Deformer (2-STD). 2-STD performs the deformation in two steps, collision avoidance step and connectivity maintenance step. It has been tested successfully on a planar robot with double integrator dynamics moving in dynamic environments.

One of the test scenarios, the robot is the green square and the dynamic obstacles are the red discs. And the plot of trajectory deformation, vertical axis is time and the green cylinder is the bounding disc of the forbidden region in ConfigurationSpaceXtime.

Publications


  • . . . . . . Vol. . No. . pp. . ed. . ()