Planning under uncertainty for safe robot exploration using Gaussian process prediction.

Area of Science:

• Robotics
• Artificial Intelligence
• Environmental Science

Background:

• Mobile robot exploration is vital for mapping unknown areas.
• Ensuring robot safety within predefined environmental condition thresholds (e.g., terrain steepness, radiation) is a significant challenge.
• Existing methods often struggle with simultaneous mapping and safe exploration.

Purpose of the Study:

• To develop a novel framework for safe exploration in unknown environments for mobile robots.
• To address two scenarios: known map with unknown safety features, and unknown map with unknown safety features.
• To enable robots to build maps while adhering to safety constraints.

Main Methods:

• Utilized Gaussian processes for predicting environmental feature values in unvisited regions.
• Developed a Markov decision process integrating Gaussian process predictions and environmental model transition probabilities.
• Incorporated the Markov decision process into an exploration algorithm prioritizing information gain, predicted safety, and proximity.

Main Results:

• The proposed framework effectively guides robots to explore new regions while maintaining safety.
• Empirical evaluations through simulations demonstrated the framework's efficacy.
• Successful application on a physical robot in an underground environment validated the approach.

Conclusions:

• The developed framework provides a robust solution for safe exploration in complex, unknown environments.
• This approach enhances robot autonomy and operational safety in challenging terrains.
• The integration of predictive modeling and decision-making processes offers a promising direction for future robotic exploration systems.