Fast Path Planning for Long-Range Planetary Roving Based on a Hierarchical Framework and Deep Reinforcement Learning

Abstract

The global path planning of planetary surface rovers is crucial for optimizing exploration benefits and system safety. For the cases of long-range roving or obstacle constraints that are time-varied, there is an urgent need to improve the computational efficiency of path planning. This paper proposes a learning-based global path planning method that outperforms conventional searching and sampling-based methods in terms of planning speed. First, a distinguishable feature map is constructed through a traversability analysis of the extraterrestrial digital elevation model. Then, considering planning efficiency and adaptability, a hierarchical framework consisting of step iteration and block iteration is designed. For the planning of each step, an end-to-end step planner named SP-ResNet is proposed that is based on deep reinforcement learning. This step planner employs a double-branch residual network for action value estimation, and is trained over a simulated DEM map collection. Comparative analyses with baselines demonstrate the prominent advantage of our method in terms of planning speed. Finally, the method is verified to be effective on real lunar terrains using CE2TMap2015.

A Review of Spatial Robotic Arm Trajectory Planning

“…They have analyzed how the SVIN formulation can improve training gradients for problems with deterministic state transition dynamics and have seen that improvement empirically on the grid world dataset. Hu et al [110] proposed a learning-based global path planning method that outperforms conventional searching and sampling-based methods in terms of planning speed. They designed a hierarchical framework consisting of step iterations and block iterations.…”

Lunar Rover Collaborated Path Planning with Artificial Potential Field-Based Heuristic on Deep Reinforcement Learning

“…To address these challenges, heuristics have been integrated into DRL frameworks to enhance search speed and circumvent local optima. Hu et al proposed SP-ResNet [18], a methodology aimed at accelerating planning speeds relative to conventional search and sampling-based techniques. Within this framework, double branches of residual networks are employed to abstract global and local obstacles, therefore constraining the search space for the DRL agent and serving as a heuristic.…”

A Deep Learning Approach to Lunar Rover Global Path Planning Using Environmental Constraints and the Rover Internal Resource Status

“…However, when addressing additional operational conditions, the concept of accelerated exploration does not necessarily equate to effective planning. A number of studies have been conducted on global path planning, employing different algorithms to address various environmental considerations: obstacle avoidance [3] (MDP); terramechanics [4,5] (Dijkstra), [6] (Reinforcement learning); sun-synchronous motion [7] (A*), [8] (Multi-speed spatiotemporal A*); terramechanics and power generation [9] (A*), [10] (Reinforcement learning); thermal condition, power generation, and terramechanics [11] (Dijkstra); uncertainty of the information [12] (RRT*); and hazard risk and collision avoidance [13] (A*), [14] (MDP), [15] (A*). These studies emphasize the importance of carefully selecting mathematical models and algorithms based on the specific purpose and constraints to be taken into account in the path planning process.…”

“…As a result, small rovers will cause immediate change in temperature and battery status in accordance with local lunar surface temperature as well as the sun position, which constantly changes over the course of the mission period. Therefore, it is essential to control when to move (timings of relocation), as well as where to move (path), to circumvent the variation in thermal and luminous conditions the rover will encounter [7][8][9][10][11].…”

“…There are fundamentally two possible ways to consider thermal and power constraints in the path planning process. One option is to use extrinsic conditions, such as lunar surface temperature and luminous environmental conditions, to determine immediate traversal/untraversal areas, such as those described in [7][8][9][10][11]. In this research, this scenario is called environment-based path search.…”