Informative Trajectory Planning Using Reinforcement Learning for Minimum-Time Exploration of Spatiotemporal Fields

Area of Science:

• Robotics
• Artificial Intelligence
• Autonomous Systems

Background:

• Existing research focuses on maximizing information gain in spatial fields.
• Efficient exploration of spatiotemporal fields with unknown distributions is crucial for autonomous systems.
• Current methods may not optimize for minimum time under information constraints.

Purpose of the Study:

• To develop a minimum-time trajectory planning method for autonomous vehicles exploring spatiotemporal fields.
• To address the challenge of efficient exploration under cumulative information constraints.
• To propose a reinforcement learning-based approach for continuous policy learning.

Main Methods:

• Modeling the problem as a Markov decision process (MDP).
• Proposing a reinforcement learning (RL) algorithm to learn a continuous planning policy.
• Designing a novel reward function using field approximations to accelerate policy learning.

Main Results:

• Proving the existence of a minimum-time trajectory under mild conditions.
• Demonstrating that the learned RL policy achieves efficient exploration.
• Showing superior performance compared to coverage planning in terms of exploration time.

Conclusions:

• The proposed RL-based trajectory planning method enables efficient spatiotemporal field exploration.
• The approach effectively balances minimizing exploration time with satisfying information constraints.
• This work advances autonomous vehicle capabilities in complex environmental surveying.