Robot movement planning for obstacle avoidance using reinforcement learning.

Area of Science:

• Robotics
• Artificial Intelligence
• Machine Learning

Background:

• Robotic arms in industrial/lab settings face challenges with obstacle avoidance in cluttered 3D spaces.
• Traditional path planning methods often struggle with local optima and efficiency in complex environments.

Purpose of the Study:

• To develop a novel reinforcement learning framework for enhanced obstacle avoidance and motion planning in robotic arms.
• To address limitations of traditional artificial potential field (APF) methods, particularly local optima and performance in dense obstacle scenarios.

Main Methods:

• A hybrid approach combining a modified artificial potential field (APF) method with the Deep Deterministic Policy Gradient (DDPG) algorithm.
• Formulation in a continuous environment to better represent real-world robotic arm operations.
• Integration of reinforcement learning factors and a tailored reward mechanism with a compensation term.

Main Results:

• The proposed framework successfully navigates complex 3D spaces, optimizing end-effector trajectories and ensuring full-body collision avoidance.
• Demonstrated mitigation of APF's local optimum issues, especially in environments with closely spaced obstacles.
• Achieved safer and more efficient obstacle avoidance with fewer steps and lower energy consumption compared to baseline models (e.g., TD3).

Conclusions:

• The novel reinforcement learning framework significantly improves robotic arm motion planning and obstacle avoidance capabilities.
• The approach offers a robust solution for complex, cluttered environments, outperforming existing methods in efficiency and safety.
• This work holds substantial potential for advancing intelligent automation in practical robotic applications.