Dynamic Target Driven Trajectory Planning using RRT

摘要

In this paper, we focus on dynamic trajectory planning for an autonomous underwater vehicle (AUV). Specifically, we are interested in planning the trajectory of an AUV as it returns to a moving recovery vessel. To aid in this task, the AUV is equipped with a passive, angle-only, sensor to enable localization of the recovery vessel. Accordingly, we present an algorithm that is capable of dynamically updating the trajectory of the AUV given measurement data from the passive sensor. Our approach is based on adapting a static trajectory planning algorithm from robotics, known as Rapidly-exploring Random Tree (RRT*), to allow for localization and tracking of a dynamic target (i.e. the recovery vessel). In contrast to dynamic programming or fixed grid trajectory planning, the RRT* offers a computationally efficient method for long-term trajectory planning with probabilistic guarantees of optimality. In this framework, we explore two options: trajectory planning based on minimising the distance to the target; and trajectory planning based on maximising the tracking accuracy of the target using an information theoretic cost. Using AUV recovery as an evaluation scenario, we analyse and evaluate the proposed trajectory planning algorithm against traditional dynamic programming methods. In particular, we consider trajectory planning in noisy and obstructed environments.