Computationally efficient and sub-optimal trajectory planning framework based on trajectory-quality growth rate...

Area of Science:

• Robotics
• Planetary Science
• Artificial Intelligence

Background:

• Planetary exploration rovers require autonomous systems with limited onboard computing power and strict power supply constraints.
• Computationally efficient algorithms are crucial for rover autonomous systems to manage processing performance effectively.

Purpose of the Study:

• To present a computationally efficient and sub-optimal trajectory planning framework for planetary exploration rovers.
• To address the trade-off between trajectory optimality and computational burden in rover navigation.

Main Methods:

• Exploited an incremental search algorithm for trajectory planning, analyzing the trajectory-quality growth rate (TQGR) to balance optimality and computational cost.
• Developed a machine learning model offline to predict the planning stop criterion based on terrain features.
• Implemented an online motion planning approach that interrupts incremental search using the predicted criterion to achieve sub-optimal trajectories with reduced computational load.

Main Results:

• The proposed framework reduced computational cost by an average of 47.6% while preserving 63.8% of trajectory optimality in simulations across diverse terrain data.
• The framework demonstrated robust performance even when the planning stop criterion prediction was not precise.

Conclusions:

• The developed trajectory planning framework offers a computationally efficient solution for planetary rover navigation under constrained resources.
• The integration of machine learning for predicting stop criteria enhances the adaptability and efficiency of autonomous rover operations in varied terrains.