All projects
Research

Autonomous Vehicle Collision Avoidance

Reinforcement Learning in a SUMO Traffic Simulation

A collision-avoidance policy for autonomous vehicles, trained with reinforcement learning inside a SUMO traffic simulation environment.

PythonSUMOReinforcement LearningTraffic Simulation

01 — Problem

What was hard about this

Collision-avoidance behaviour can't be learned safely on real roads, and hand-written rules struggle with the long tail of traffic situations. Training needs an environment that generates those situations repeatedly and cheaply.

02 — Solution

How it works

Carried out as a Research Assistant at Nirma University under Prof. Anuja Nair. Built the traffic simulation environment in SUMO to generate the interaction scenarios, then trained a reinforcement-learning policy against it to produce the avoidance behaviour.

03 — Impact

What shipped

  • Traffic simulation environment built in SUMO to generate collision scenarios repeatably
  • Collision-avoidance policy learned through reinforcement learning rather than hand-written rules
  • No benchmark figures are published here — the evaluation was in-simulation and I don't have a protocol I'd stand behind
Back to all projectsWant to talk about how this would fit your team?