Safe contact-based robot active search using Bayesian optimization and control barrier functions

Area of Science:

• Robotics
• Artificial Intelligence
• Medical Imaging

Background:

• Safe exploration is crucial for robots in uncertain environments to prevent damage.
• Autonomous ultrasound scanning requires precise navigation and controlled interaction.

Purpose of the Study:

• To develop a safe active search method for autonomous robots.
• To ensure patient safety and improve image quality in medical robotics applications.

Main Methods:

• Utilizing Bayesian optimization with an informative continuous expected improvement acquisition function.
• Implementing exponential control barrier functions to bound robot-environment contact forces.
• Leveraging the derivative of force in the contact model for enhanced robustness.

Main Results:

• Achieved 92% success rate in limited time (under 60s) for active search, a 13% improvement over standard expected improvement.
• Successfully limited robot contact forces to under 5 N, even with inaccurate contact boundary specifications.
• First demonstration of safe active search on an autonomous robot for rheumatoid arthritis ultrasound scanning.

Conclusions:

• The proposed method enables safe and efficient autonomous robotic operations in uncertain environments.
• Control barrier functions are effective in ensuring safety and robustness in medical robotics.
• This approach significantly improves active search performance and guarantees patient safety.