Research map / Reinforcement learning / Visual and representation RL
SafeLight: A Reinforcement Learning Method toward Collision-Free Traffic Signal Control
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add safety-rule action filtering | 3DQN · Synthetic dataset, cyclic actions | collisions: 93.94% fewer collisions | Helped |
| Add safety-rule action filtering | 3DQN · Real-world dataset | collisions: 86% fewer collisions | Helped |
| Add safety-based reward shaping | 3DQN · Synthetic setting | collisions: 99.32% improvement in safety | Helped |
| Add a safety module to the RL method | Backbone RL (3DQN) · Synthetic and real-world settings | collisions: over 99 percent reduction in collisions | Helped |
| Add a safety module to the RL method | Backbone RL (3DQN) · Compared with Fixed-time control | average waiting time: about 30 percent lower average waiting time | Helped |
| Add an additional safety feature to the state representation (ablation) | SafeLight-R or SafeLight-Loss · Three evaluation settings | overall performance | No clear effect |
| Use a synthetic weighted-Q objective instead of the backbone objective | 3DQN · Compared with Backbone RL (3DQN) | average waiting time | Hurt |
| Use synthetic reward shaping | 3DQN · Cyclic versus acyclic action settings | collisions | Mixed |
| Use acyclic rather than cyclic actions | Traffic signal controllers · Different action-space settings | collisions | Hurt |
| Add a safety penalty to the reward | IPPO · Synthetic intersection | collisions: 24.5 -> 0.0 | Helped |
| Add a safety objective to the loss | IPPO · Synthetic intersection | average delay: 10.66 -> 13.07 | Hurt |
Closest papers by meaning
- RobustLight: Improving Robustness via Diffusion Reinforcement Learning for Traffic Signal Control (ICML 2025 · similarity 0.89)
- π-Light: Programmatic Interpretable Reinforcement Learning for Resource-Limited Traffic Signal Control (AAAI 2024 · similarity 0.88)
- VLMLight: Safety-Critical Traffic Signal Control via Vision-Language Meta-Control and Dual-Branch Reasoning Architecture (NeurIPS 2025 · similarity 0.88)
- DiffLight: A Partial Rewards Conditioned Diffusion Model for Traffic Signal Control with Missing Data (NeurIPS 2024 · similarity 0.87)
- Latent Safety-Constrained Policy Approach for Safe Offline Reinforcement Learning (ICLR 2025 · similarity 0.85)
- IntersectionZoo: Eco-driving for Benchmarking Multi-Agent Contextual Reinforcement Learning (ICLR 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.