Making cars drive themselves.
AV@B is a UC Berkeley RSO working on making self-driving cars and bringing them to campus.
Built on
ROS 2
NVIDIA
PyTorch
Python
C++
OpenCV
NumPy
Gazebo
Docker
Ubuntu
CMake
Jupyter
Git
GitHub
AV@B is one autonomy stack and the people building it: students from machine learning, robotics, controls and hardware who put their code on real vehicles.
Read our write-ups ›How a car drives itself
Berkeley Driver runs the same four steps many times a second, on every vehicle we build.
Sense
Cameras, lidar, radar, IMU and GPS read the world around the car.
01Perceive
Detection, depth and occupancy turn raw data into cars, lanes and people.
02Plan
Prediction and planning choose a safe path and speed through traffic.
03Act
Control turns the plan into steering, throttle and braking.
04Our engineers are engineers and researchers from
Berkeley AI Research
Overshoot
Y Combinator
Givefront
Owner.com
NYUFour divisions, one goal.
Research
Papers, models and datasets on driving, perception and planning.
Research ›Racing
Small-scale cars, then go-karts, then full-size race cars.
Racing ›Mobility
Self-driving vehicles for Berkeley, starting with a golf cart.
Mobility ›Applied
Paid autonomy and robotics projects for startups and labs.
Applied ›What we're building
Berkeley Driver
Our autonomous driving stack. One codebase for perception, prediction, planning, localization, control, simulation and deployment.
- Runs on every vehicle
- Shared by all four divisions
Autonomous golf cart
Our first vehicle you can sit in. A drive-by-wire cart for testing the stack in controlled areas before larger vehicles.
- Drive-by-wire
- Human scale
Autonomous racing
The stack at full speed. Fast autonomous cars built to race other universities and autonomous racing teams.
- Small-scale cars
- Go-karts
- Full-size race cars
Write-ups
How our stack, vehicles and research fit together. Papers will be added here as they're published.
Berkeley Driver: one stack for every vehicle
Read ›Our autonomous driving stack. One codebase for perception, prediction, planning, localization, control, simulation and deployment.
The autonomous golf cart
Read ›Our first human-scale vehicle. A drive-by-wire cart for testing the stack in controlled areas before larger deployments.
Autonomous racing: the stack at full speed
Read ›Fast autonomous cars built to race other universities and autonomous racing teams, moving from small-scale cars to full-size race cars.
What AV@B Research works on
Read ›Ten topics across embodied AI, driving, perception and planning, and how a research project turns into papers, models, datasets and code.
One codebase, four proving grounds
Read ›How Research, Racing, Mobility and Applied feed each other through a shared driving stack, and where the money goes.
Toward an autonomous service on campus
Read ›The long-term Mobility goal: a safe, approved, geofenced autonomous service for the Berkeley community, and the five phases to get there.
Small teams that ship.
We keep teams small enough that every member owns a real piece of a system. Open to Berkeley students in any major, and you don't need autonomy experience.
Real vehicles
Your code runs on robots, a golf cart and race cars.
Ownership
A system that is yours to design, test and ship.
Industry work
Paid projects for companies through AV@B Applied.