TOWARDS COGNITIVE STEERING BEHAVIOURS FOR TWO-WHEELED ROBOTS

We present a two-layer architecture for two-wheeled robots trajectory planning. This architecture can be used to describe steering behaviours and to generate candidate trajectories that will be evaluated by a higher-level layer before choosing which one will be followed. The higher layer uses a TAEMS tree to describe the current robot goal and its decomposition into alternative steering behaviours. The lower layer uses the DKP trajectory planner to grow a tree of spline trajectories that respect the kinematic constraints of the problem, such as linear/angular speed limits or obstacle avoidance. The two layers closely interact, allowing the two trees to grow simultaneously: the TAEMS tree nodes contain steering parameters used by DKP to generate its branches, and points reached in DKP tree nodes are used to trigger events that generate new subtrees in the TAEMS tree. We give two illustrative examples: (1) generation and evaluation of trajectories on a Voronoi-based roadmap and (2) overtaking behaviour in a road-like environment.