In the pursuit of efficient traffic flow, Adaptive Traffic Signal Control (ATSC) technologies continuously modify signal timings in real time to rapidly respond to traffic changes, optimize timing decisions, and minimize delays and negative externalities.
Monte Carlo Lights is a novel ATSC approach that combines state-of-the-art reinforcement learning and Monte Carlo tree search to optimize traffic flow.
In recent years, AI-based Deep Reinforcement Learning (DRL) optimization approaches, have proven their superiority to older model-based approaches in traffic signal control. While model-based ATSC systems rely on simple models and coarser observations (e.g. traffic queues), DRL approaches ingest rich observations (e.g. position and speed of all vehicles within the detection area) and learn directly from interacting with traffic. Though these model-free DRL agents work very well when trained well, they may fail to generalize in the real-world to substantially different intersection layouts and traffic demands, requiring re-training in the new environments. We overcome these challenges with model-based RL and Monte Carlo Tree search techniques.
Our hybrid approach:
- Delivers demand-responsive and self-learning optimal control decisions
- Requires little up-front training data without the need for retraining
- Handles different intersection layouts at scale
OPPORTUNITY
The market for automated traffic control applications was valued at $26.8 billion in 2019 and is expected to reach $64.0 billion by 2025, a CAGR of 15.2% (BCC Research). As urban populations grow, efficient transportation systems become critical for quality of life, economic productivity, and the environment. Smart traffic systems can yield cost-effective benefits without major upgrades to existing infrastructure.
The Monte Carlo Lights ATSC can provide benefits such as:
- Reducing congestion by smoothing and prioritizing traffic flows in response to real time demand
- Lower pollution levels by reducing stop-start driving
- Enabling a more effective response to traffic incidents
- Prioritizing public transportation vehicles over private cars
STATUS
- Patent applications filed in US and Canada
- Seeking partners for licensing
