A MDPs-Based Dynamic Path Planning in Unknown Environments for Hopping Locomotion

Abstract

Hopping robots, or “hoppers”, are promising explorers capable of navigating rough terrain such as disaster areas and celestial environments. For example, rovers exploring planetary surfaces need to minimise the risk of failure and maximise the acquisition of information about their environment. Hopping locomotion in such environments is inherently uncertain because the details of the environment are largely unknown. As a result, path planning algorithms must account for these uncertainties in order to effectively traverse these environments. This study presents a novel hopping path planning algorithm for uncertain environments. The proposed algorithm uses Markov Decision Processes (MDPs) to compute motion uncertainties and subsequently generates optimal actions for all states according to the terrain conditions and mission requirements. In addition, the proposed algorithm incorporates a perception method using hopping locomotion features, which enables dynamic path generation. The proposed algorithm is evaluated through simulations in three different environments, which demonstrate that the hopper can achieve its goals with a 98% success rate on hard ground and heterogeneous terrain, and over 80% success rate on sandy terrain, using the proposed algorithm. Furthermore, the robustness of the proposed algorithm is validated through comparison with a greedy algorithm using the same payoff function, which shows that the proposed algorithm achieves 20 times higher success rate and 38.7% lower average number of steps than the greedy algorithm in the best case scenario.

Keywords

• Motion and path planning
• optimization and optimal control
• space robotics