HELICOPTER PATH PLANNING USING U-NET INFORMED BIASED SAMPLING

Abstract

To increase the safety of helicopter operations during missions in close proximity to obstacles under degraded visual conditions, assistance systems such as obstacle-avoiding path planning algorithms can be used. However, conventional planners may take a long time to find a suitable solution, respectively to converge to an optimum, when used in high dimensional state spaces. This paper describes a method to enhance the convergence time of sampling-based nonholonomic path planning algorithms. To accomplish this, the sampling distribution is biased towards the most promising regions. These distributions are approximated based on the obstacle scenery as well as start and goal state using a U-NET and artificial potential fields. Experiments were conducted with the proposed technique implemented into a BIT*-algorithm - which demonstrated a reduction in average path cost - and into an RRT-algorithm, resulting in improvements in the overall solution cost. This paper applies the described optimization method to planning problems in three dimensions, considering simplified helicopter kinematics, using an expanded Dubins state space