A learning-based semi-autonomous controller for robotic exploration of unknown disaster scenes while searching for...

Area of Science:

• Robotics
• Artificial Intelligence
• Disaster Response

Background:

• Teleoperation and fully autonomous control have limitations in disaster environments.
• Semi-autonomous control allows human-robot collaboration for complex tasks like navigation and exploration.
• Urban Search and Rescue (USAR) operations require robots capable of handling cluttered and unknown terrains.

Purpose of the Study:

• To present a novel hierarchical reinforcement learning (HRL)-based semi-autonomous control architecture for rescue robots.
• To enable rescue robots to learn from experience and improve performance in exploring unknown disaster scenes.
• To enhance robot exploration capabilities using a direction-based technique for region and rubble classification.

Main Methods:

• Developed a unique hierarchical reinforcement learning (HRL) control architecture.
• Integrated a direction-based exploration technique for region and rubble pile classification.
• Validated the controller through simulations and physical experiments in USAR-like environments.

Main Results:

• The proposed HRL-based semi-autonomous controller demonstrated robustness in unknown, cluttered USAR environments.
• The controller facilitated continuous learning and performance improvement for robot exploration.
• The direction-based exploration technique effectively expanded the robot's search area.

Conclusions:

• The HRL-based semi-autonomous control architecture is effective for rescue robots in challenging USAR scenarios.
• This approach enhances robot autonomy and adaptability in disaster response operations.
• The system shows promise for improving the efficiency and safety of urban search and rescue missions.