RL-PID rover-arm simulator
A factory-style rover-arm simulator where reinforcement learning supervises classical PID motion.
Built the simulator structure, controller comparisons and safety-layer experiments.
PID vs rules vs RL supervision
- Built a factory-style rover-arm simulator where RL supervises classical PID motion.
- Compared controllers and safety-layer experiments on shared metrics.
- Result / current state
- The design compares PID-only, rule-based safety and RL-supervised control on shared metrics.
- Technologies
- Python
- PyBullet
- Gymnasium
- Stable-Baselines3
What it does
The simulator moves a rover-mounted arm toward a target while avoiding a moving human-like obstacle in a factory-style workspace.
My contribution
I built the simulation structure and controller comparisons for PID-only, rule-based safety and a DQN supervisor.
Technical highlight
PID handles smooth low-level motion while high-level actions choose whether to drive, slow, turn, wait, reverse or move the arm to a safe pose. A safety layer enforces hard collision and distance limits.
Results and status
This is an MVP in progress. It is a simulator and controller study, distinct from the Dr. Robot rover-arm work, with no hardware deployment claim.