Improved PPO Optimization for Robotic Arm Grasping Trajectory Planning and Real-Robot Migration.

Area of Science:

• Robotics and Artificial Intelligence
• Machine Learning
• Control Systems

Background:

• Robotic manipulation in unstructured environments faces challenges like local optima and convergence issues.
• Real-time interaction and adaptability are crucial for industrial automation.

Purpose of the Study:

• To develop a hybrid reinforcement learning framework for precise, collision-free robotic arm trajectory planning.
• To enhance the grasping of randomly appearing objects amidst dynamic obstacles.

Main Methods:

• A hybrid approach combining Simulated Annealing (SA) with Proximal Policy Optimization (PPO).
• A probabilistically enhanced simulation environment with a 20% obstacle generation rate.
• An optimized state-action space with 12-dimensional environment coding and 6-DoF joint control.

Main Results:

• Achieved a 6.52% increase in success rate (98% vs. 92%) compared to baseline PPO.
• Reduced steps per set by 7.14%.
• Validated robust simulation-to-reality transfer on an AUBO-i5 robotic arm.

Conclusions:

• The SA-PPO algorithm effectively balances exploration and convergence for adaptive robot manipulation.
• This research establishes a new paradigm for real-time robotic response to environmental uncertainty in industrial settings.