Researchers have introduced Sim2Signal, a new benchmark designed to systematically measure and evaluate methods for bridging the "Sim-to-Real gap" in traffic signal control using reinforcement learning. This gap, where policies trained in simulations fail in real-world deployment, is decomposed into observation, action, transition, and reward components, each induced in isolation. Experiments across 33 gap settings and 10 calibrated networks revealed that direct transfer consistently degrades performance, and mitigation effectiveness is highly dependent on the specific network and gap type. The study found that methods estimating gap changes were generally more effective than domain randomization techniques. AI
IMPACT This benchmark could accelerate the development and deployment of more robust AI systems for real-world traffic control by providing a standardized way to measure and address the Sim-to-Real gap.
RANK_REASON The cluster contains a research paper introducing a new benchmark for evaluating AI methods. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- DagsHub
- Domain Randomization
- Hugging Face
- Markov decision process
- reinforcement learning
- Sim2Signal
- Sim-to-Real Gap
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
