Real-time Trajectory Generation for Collision Avoidance with Obstacle Uncertainty
Chi Kin Lai, James Ferris Whidborne · 2011
This paper first presents the integration of trajectory generation into an autonomous collision avoidance system and then proposes a systematic update scheme for trajectory generation so that real-time feedback can be obtained to compensate for obstacle uncertainty. Driven by the promise of optimization-based methods for collision avoidance, the real-time update scheme attempts to alleviate two of the well-known problems that are intrinsic to optimization-based methods: namely no guarantee for convergence and dependence on predictions (models). Both the not-converged solutions and prediction uncertainty may endanger the host aircraft by leading it to a possibly infeasible trajectory. This paper demonstrates that the proposed approach is able to accommodate some degree of uncertainty by utilizing real-time feedback and to handle not-converged situations by appending an onboard backup trajectory to the avoidance trajectory at every update step. These are illustrated with two examples and the effects of obstacle uncertainty and not-converged solutions on the vehicle safety are discussed.