Energy Optimized Green Light Assist in Varying Traffic Scenarios Using Reinforcement Learning
summary
A speed-advice system compares a deterministic timing-based controller and a PPO RL agent (MLP actor-critic) in SUMO on a 5 km single‑lane approach with randomized signals (green 10–20 s, red 25–35 s, yellow 3 s) and only electric cars; state includes vehicle speed, distance to light, signal timings, radar gap and cumulative energy. Results show the RL agent cuts energy by about 22.83% energy drop at 20 vehicles with ~+1.73% travel time, trained over 1,000 episodes per config; limitations: single intersection, single lane, full compliance assumptions.
AI-written summary · about
chapters (15)
- 0:00Green light assist research intro
- 1:10Traffic light optimization approach
- 2:36Vehicle-to-X communication setup
- 4:42Deterministic speed recommendation
- 6:30Reinforcement learning state design
- 8:35Energy and time reward components
- 10:21Evaluation metrics and testing
- 11:46Deterministic algorithm results
- 12:07Reinforcement learning performance across traffic volumes
- 13:30Speed vs energy consumption analysis
- 14:58RL energy variance reduction
- 15:40RL agent behavior demo
- 17:06Simulation limitations
- 18:54Conclusion and method selection
- 19:16Q&A
related talks
21mChallenges in Reward Design for Reinforcement Learning-based Traffic Signal Control
15mA Comparison of Reinforcement Learning Agents Applied to Traffic Signal Optimization
13mAUTO: Adaptive Congestion Control Based on Multi-Objective Reinforcement Learning...
11mUsing Deep Reinforcement Learning to Coordinate Multi-Modal Journey Planning with Limited ...
13mWIP: Learning Adversarial Attacks on Adaptive Traffic Signal Control Systems...
19mTestbed for Development of AI Solutions to Boost Operational Efficiency in Transportation Networks
20mAI on the Fast Lane: Smart Traffic Systems to Unclog our Cities
18mMaximizing Luck in Reinforcement Learning