Approche cognitive pour la planification de trajectoire sous contraintes. (A cognitive approach for kinodynamic trajectory planning)

A cognitive approach for kinodynamic trajectory planning In this thesis, I present two new contributions. First, I provide DKP: a sample-based approach for trajectory planning. DKP uses a selection/propagation architecture to build an exploration tree in the reachable parts of the environment, guided in an A manner. Solutions are quadratic spline trajectories that are immediately executable by two-wheeled robots. The propagation level builds a parameter space which contains all the values of the free parameters in the solution. It is represented as a surface containing all local solutions which respect kinodynamic constraints: speed, acceleration, obstacle avoidance. . . Then, a search is applied on the parameter space using an optimization criterion. DKP is deterministic: every result produced by DKP may be repeated. Second, this control is used to define steering behaviors. These are expressed within a steering tree: every behavior acts on the way the exploration tree built by DKP progresses in the environment. Steering behaviors are applied according to the explored part. Thereby, TÆMS is used to describe the steering behaviors and to evaluate the solutions. To sum up, my cognitive approach takes advantage on the common building of a steering tree and an exploration tree which validates respect of constraints: thus, we get a link between classing planning and trajectory planning under constraints.