Search code, repositories, users, issues, pull requests...
This project is a custom OpenAI Gym-compatible environment designed for simulating a UR5 robotic arm with a Robotiq 2F-85 gripper using PyBullet. The environment enables training and testing reinforcement learning algorithms for a basic pick-and-place task in a 3D physics-based simulation.
The agent learns to move the end-effector to the correct [x, y] position above a cube and perform a successful grasp using inverse kinematics and simulated gripper control.
Among these, SAC consistently achieved higher success rates and faster convergence. It was more stable and sample-efficient for this continuous control task.
Below is a comparison of the training reward progress for each algorithm. SAC shows the best performance in terms of higher rewards and faster convergence.
