Research
UC Santa Cruz · May 2025 – Jun 2026
Prioritized Level Replay in RL Driving Environments
- Prioritized level replay as a curriculum for PPO driving agents
- Highway scenarios generated with Scenic, run in MetaDrive
- Beat plain PPO in 21 of 30 seeds (p < 0.05)
About
Demonstrates that prioritized level replay, an automated curriculum learning method for RL, improves PPO performance in highway driving scenarios with reactive traffic. Scenes are sampled from Scenic scenario specifications into MetaDrive, scored by positive value loss, and resampled by rank. In a controlled 30-seed comparison, PLR-trained agents beat the unmodified PPO baseline in 21 of 30 seeds (sign test, p < 0.05).
- Python
- Reinforcement Learning
- PPO
- Autonomous Driving