Learning to Avoid Obstacles With Minimal Intervention Control.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Programming by demonstration (PbD) enables robots to acquire motor skills from human teachers.
• Traditional imitation learning often focuses on single spaces (Cartesian or joint), limiting applicability.
• Hybrid imitation learning addresses tasks requiring simultaneous control in both spaces.

Purpose of the Study:

• To develop a method for hybrid imitation learning that incorporates obstacle avoidance.
• To enable robots to generalize learned skills to dynamic environments with obstacles.
• To ensure safety during human-robot interaction while navigating obstacles.

Main Methods:

• Designing potential fields for effective obstacle bypassing.
• Ensuring joint limits are respected during obstacle avoidance trajectory adjustments.
• Developing control commands for safe human-robot interaction.

Main Results:

• The proposed method allows robots to learn and execute skills in the presence of obstacles.
• Validation demonstrated feasibility and safety in both simulated and real-world experiments.
• The robot successfully generalized observed skills to new situations with obstacles.

Conclusions:

• The developed approach enhances hybrid imitation learning by integrating robust obstacle avoidance.
• This research contributes to more adaptable and safer robot skill acquisition in complex environments.
• The method is effective for real-world applications, as shown with the iCub humanoid robot.