Path Planning & Control of a Nonholonomic Autonomous Robotic System for Docking
By Chadwick Allen Sylvester
December 2003
Chair: Gloria J. Wiens
Cochair: Norman G. Fitz-Coy
Major Department: Mechanical and Aerospace Engineering
Autonomous maneuvering can be a difficult and broad issue for multiple robots performing various tasks. In the coordination control of multiple robots, compatible trajectories for each robot are essential. This is especially critical for nonholonomic robots performing collaborative tasks. To ensure proper movements performed by the robots, trajectories in time must first be achieved to guarantee efficient, non-impeding traversals. The applications where this is pertinent are transportation, surveying, formation and following. Position feedback and path planning are two major areas of study that have developed from these endeavors.
In this thesis, an architecture is presented and implemented that uses classical control methods to track the position of two robots in a workspace. The goal of their maneuvers is to have the parallel-steered robots traverse a predetermined trajectory and result in a configuration that allows for front-to-front docking. A non-model based approach is applied in the development of a path planning algorithm for tracking control subject to physical constraints of the robot’s maneuverability. A framework for creating a trajectory via five constraints applied to a fourth order polynomial is presented and implemented. The predetermined trajectory is obtained using a simplified polynomial bisection search technique for finding optimal polynomial coefficients. The polynomial based trajectory also ensures that the minimum turning radius of each vehicle is not violated for any part on the trajectory. The controller incorporates the speed and steering to achieve the desired position and heading of the robots as well as ensure that each vehicle traverses the prescribed trajectory to a docking location.
Using an off-board 3D camera system and robots equipped with LEDs, this algorithm is implemented in a tracking controller for two parallel-steered robots required to maneuver and dock with one another. Both the speed and steering controls require delicate tuning of gains for accurate responses. Various experiments are performed while modifying the initial poses of the vehicles. Tuning methods are used as opposed to a model-based approach because the inaccuracies of the steering mechanism and drive gearbox would negate the advantages achieved by the dynamical model obtained.
The outcome of this research is the fundamental path planning architecture for a hierarchical controller in which the robots’ pose is sent to the collective intelligence of the team where the decisions are generated to plan trajectories and control the robots in coordinated tandem maneuvers. The corresponding position measurement system is an overhead, off board system, which allows for collection of global information about the positions of the robots and their goals.