Evaluation of Sampling-Based Optimizing Planners for Outdoor Robot Navigation

TL;DR

This paper systematically evaluates the performance of various Sampling-Based Optimal (SBO) planners from OMPL for outdoor mobile robot navigation in 3D environments, highlighting the benefits of parallel computing for faster convergence.

Contribution

It provides a comprehensive comparison of SBO planners in outdoor navigation scenarios, using randomized problem generation and multiple metrics, emphasizing the importance of parallel computing.

Findings

Parallel computing improves convergence rates of SBO planners.

Certain SBO planners consistently outperform others across different environments.

Evaluation framework aids in selecting suitable planners for outdoor robot navigation.

Abstract

Sampling-Based Optimal(SBO) path planning has been mainly used for robotic arm manipulation tasks. Several research works have been carried out in order to evaluate performances of various SBO planners for arm manipulation. However, not much of work is available that highlights performances of SBO planners in context of mobile robot navigation in outdoor 3D environments. This paper evaluates performances of major SBO planners in Open Motion Planning Library(OMPL) for that purpose. Due to large number of existing SBO planners, experimenting and selecting a proper planner for a planning problem can be burdensome and ambiguous. SBO planner's probabilistic nature can also add a bias to this procedure. To address this, we evaluate performances of all available SBO planners in OMPL with a randomized planning problem generation method iteratively. Evaluations are done in various state spaces suiting for different differential constraints of mobile robots. The planning setups are focused for navigation of mobile robots in outdoor environments. The outdoor environment representation is done with prebuilt OctoMaps, collision checks are performed between a 3D box representing robot body and OctoMap for validation of sampled states. Several evaluation metrics such as resulting path's length, smoothness and status of acquired final solutions are selected. According to selected metrics, performances from different SBO planners are presented comparatively. Experimental results shows the significance of parallel computing towards quicker convergence rates for optimal solutions. Several SBO methods that takes advantage of parallel computing produced better results consistently in all state spaces for different planning inquiries.