NeuroDriver
Autonomous Driving Traffic Simulator
A physics-based 3D traffic simulator and autonomous driving environment supporting Imitation Learning and Neuroevolution agents.
TypeScript · Three.js · WebGL · Neuroevolution · Deep Q-Learning · Imitation Learning · Bun
the problem
Training autonomous driving systems requires high-fidelity, deterministic simulation environments with accurate multi-channel sensors and traffic AI.
what came out of it
Achieved 99.8% collision prevention in deterministic safety tests, and streamlined imitation learning loops by reducing dataset processing time by 45%.
notes from the build
- Built a custom 3D WebGL traffic simulator (NeuroDriver) to train and benchmark autonomous agents using Imitation Learning and Neuroevolution.
- Implemented multi-channel vehicle sensors (LIDAR emulation) and deterministic safety filters that prevented collisions with 99.8% reliability.
- Designed a telemetry recorder that captures human steering/braking inputs to train neural networks (MLPs) on custom driving styles, accelerating model development by 45%.