Adaptive path planning for intelligent vehicle in dynamic and uncertain environments

This thesis explores the advancement of path planning and operational strategies for mobile robots and UAVs in dynamic, uncertain and challenging environments, with a focus on intelligent systems that contribute to human welfare. The first three works address the challenges of autonomous mobile robot navigation in human-populated environments, emphasizing crowd-aware and socially-adaptive path planning. The objective is to enable intelligent agents to operate safely and efficiently without causing discomfort to human society.

The initial work proposes a grid-based congestion-aware path planning method using crowd density map, which is generated by a proposed Partially Updated Memory (PUM) model, to account for spatial-temporal crowd anomalies. Building on this foundation, the second work propose a Probability-related Partially Updated Memory (PPUM) model with Receding Horizon Optimization (RHO) for path planning, aims to improve robots' adaptability in uncertain environments. The third work then extends 2D path planning in spatial-temporal domain by considering the potential interactions between predicted crowd distribution and the planned route, which improves navigation efficiency and social awareness through adapting the planned route to future crowd dynamics in a high-level decision making stage.

Expanding on the theme of developing adaptive path planning for intelligent vehicles, the thesis transitions from mobile robots (treated as ground vehicles) to UAVs, further focusing on operations in adverse wilderness environments. The fourth study adapts and enhances the path planning techniques initially developed for mobile robots, extending them to multi-UAV systems for Wilderness Search and Rescue (WiSAR). It introduces a smart agent-based probability model that estimates the possible locations of a lost person over time, based on terrain features. The receding horizon planning method, previously proposed for robot navigation in crowded environments, is modified for search-based planning, enabling the dynamic prioritization of search areas based on the evolving probability distribution. This framework significantly improves search efficiency, and in this application, directly contributes to human welfare by aiding in the rapid location of lost persons. Finally, the UAV landing strategy, presented as a practical extension work in Appendix, improves mission completion by ensuring safe and robust landings on a mobile base station in wildness. This work builds a practical foundation for implementing the proposed multi-UAV framework in real-world scenarios, leveraging transferable techniques validated in the landing application, including communication mechanisms between the UGV and UAVs and applied control strategies.