Greedy BIT* (GBIT*): Greedy Search Policy for Sampling-Based Optimal Planning with a Faster Initial Solution and Convergence

This paper presents Greedy Batch Informed Trees (GBIT*), a greedy version of Batch Informed Trees (BIT*) and Advanced Batch Informed Trees (ABIT*) with a greedy search policy inspired by RRT-Connect. BIT* and ABIT* use an edge queue ordered by the (inflated) potential path cost to find the best next edge to process. GBIT* builds on ABIT* by adding another preferential way, which is defined by the greedy search policy, to choose the next edge to process. Otherwise, it will follow ABIT*’s method. The greedy search policy guides the search moving forward greedily and towards the goal, which can make it faster to find the initial solution. An earlier initial solution can lead to a faster upper bound to define the informed set and start the convergence process earlier. The experiment results show that in different maps, GBIT* can find an initial solution faster than any other sampling-based asymptotically optimal planners, as well as RRT-Connect in most cases.