Autonomous racing: the stack at full speed
Project
TL;DR
AV@B Racing builds autonomous systems for competition, where performance is measurable and failure is obvious.
We start with small-scale cars, move to go-karts, and work toward full-size race cars.
Why race
A race produces a lap time and a finishing position. That makes it the clearest test of whether a change to perception or planning helped, and it finds the failures that slower testing hides.
What racing asks of the stack
- Perception at high speed
- Localization at racing pace
- Predicting what opponents will do
- Planning in real time
- Control at the edge of grip
- Inference with almost no latency
The path up
Small-scale cars → go-karts → full-size race cars
We race to win. What survives the track goes back into Berkeley Driver for every other vehicle.