Learning Environmental Field Exploration with Computationally Constrained Underwater Robots: Gaussian Processes Meet...

Area of Science:

• Robotics
• Environmental Science
• Computer Science

Background:

• Autonomous exploration by underwater robots is crucial for environmental field mapping.
• Limited underwater communication necessitates onboard computation for exploration algorithms.
• Reducing computational cost is vital for micro underwater robot teams.

Purpose of the Study:

• To develop a computationally efficient algorithm for autonomous underwater environmental field exploration.
• To maximize information gain by integrating an information-metric into path planning.
• To address the challenge of onboard computation for micro underwater robot fleets.

Main Methods:

• Utilizing field belief models based on Gaussian Processes (e.g., Gaussian Markov random fields, Kalman regression).
• Employing weighted shape functions to integrate continuous field observations.
• Implementing path planning via stochastic optimal control with path integrals.

Main Results:

• Achieved constant computational cost over time for field estimation.
• Developed belief models functioning as information-theoretic value functions.
• Demonstrated the algorithm's efficiency in simulations, including stationary spatio-temporal fields.

Conclusions:

• The proposed algorithm offers a computationally efficient solution for autonomous underwater exploration.
• Gaussian Process-based belief models with weighted shape functions enhance field estimation and path planning.
• This approach is suitable for in-field implementations on micro underwater robot teams.