Highly optimal TSC systems are not lightyears away. This is compliant and auditable RL-based signal control.
- Modeled vehicle wait
- 48.3% lower
- Modeled pedestrian wait
- 68.5% lower
- Idling CO₂
- 24.5% lower
An illustrative four-way signalised junction, seen from above, running a fourteen second cycle. The major road holds green for the first 43% of the cycle and then releases on amber; the minor road then runs green, with a pedestrian crossing phase, before the major road returns. Vehicles queue behind their stop bars and cross only on their own green. Beside the junction, a panel shows the RL model and a decision record cycling through four things: the state observed (density 0.42 vehicles per metre, flow 612 vehicles per hour, south queue six vehicles), the action the model selected (a six second extension of the north–south green, chosen from a bounded action set), the fixed-time plan it replaced (a thirty second fixed phase that would have held regardless of the queue), and the explanation written to the log. The figures and the junction are illustrative and simulated, not measured at a live intersection.
Real roads don’t behave like a simulator
Demand shifts, sensors miss things, and people on foot, on bikes, and in cars do things nobody predicted. A model that runs public infrastructure needs hard constraints and human oversight, and someone has to be able to ask it why it made a call.
Explore our point of view →Four rules we build by
Read the whole junction
The system takes in vehicle flow, pedestrian movement, signal phase, time, and how demand is changing.
Stay inside the rules
It adapts only within the operational constraints it has been given.
Show the work
Every decision should leave a record that the people responsible for the road can inspect.
Build on what is there
It is modular, so it can improve the infrastructure already in place.
Traditional RL is a blackbox, so we fixed that
CObRA™ turns a model action into an inspectable record: what the system saw, what it chose, an explanation of why it chose its decision, and how confidently that choice can be explained.
Inspect CObRA™ ↗Better flow starts with better feedback
We are building tools for the people who keep cities moving. Bring us the intersection that keeps you up at night.
Start a conversation ↗