The challenge of finding an optimized and reliable path dates back to emersion of mobile robots. Several approaches have been developed that have partially answered this need.

Satisfying results in previous implementations has led to an increased utilization of sampling-based motion planning algorithms in recent years, especially in high degrees of freedom (DOF), fast evolving environments. Another advantage of these algorithms is their probabilistic completeness that guarantees delivery of a path in sufficient time, if one exists.

On the other hand, sampling based motion planners leave no comment on safety of the planned path. This paper suggests biasing the Rapidly-exploring Random Trees (RRTs), with the outcome of a safety evaluation, which affects the probability of choosing a random point in the sampling phase of the RRT algorithm, to increase the chance of safer outcomes.

By parallelizing this algorithm and multifold execution of it on the Graphics Processing Unit (GPU) with various probabilities for moving to a safer state, a near-optimal solution is obtained.

Here is our paper: A method for real-time safe navigation in noisy environments