autowarefoundation/auto_fsd

★ 14PythonAudience · researcherComplexity · 5/5LicenseSetup · hard

Mindmap

mindmap
  root((AutoFSD))
    What it does
      End-to-end driving
      Camera-only input
      No HD maps needed
    Use cases
      Robotaxi driverless
      L2 plus plus assist
      Highway and city
    Tech
      Python
      PyTorch
      Optional LIDAR RADAR
    Audience
      AV researchers
      Autoware developers

mindmap root((AutoFSD)) What it does End-to-end driving Camera-only input No HD maps needed Use cases Robotaxi driverless L2 plus plus assist Highway and city Tech Python PyTorch Optional LIDAR RADAR Audience AV researchers Autoware developers

Click or tap to explore — scroll the page freely

Things people build with this

USE CASE 1

Train a camera-only self-driving model for highway or city roads without needing pre-built HD maps

USE CASE 2

Combine with LIDAR or RADAR sensors to build a fully driverless robotaxi system

USE CASE 3

Add L2++ hands-free point-to-point navigation to a vehicle while keeping a human available to take over

USE CASE 4

Extend or compare against the traditional Autoware stack with an end-to-end neural network approach

Tech stack

PythonPyTorch

Getting it running

Difficulty · hard Time to first run · 1day+

Requires a GPU-enabled PyTorch environment, the README is early-stage and does not yet document the full training or inference workflow.

Use, modify, and distribute freely for any purpose including commercial, with attribution. Apache 2.0 also protects contributors from patent claims.

In plain English

AutoFSD is an open-source AI model for self-driving vehicles, released by the Autoware Foundation. The foundation also maintains Autoware, a well-known open-source autonomous driving software stack, and this project extends that work with an end-to-end neural network approach. The model is designed to drive on highways, arterial roads, and city streets using only cameras as sensors, without needing pre-built high-definition maps of the area. The "end-to-end" part of the name means that the AI takes raw camera images as input and outputs driving decisions directly, rather than going through a series of separate steps like detecting objects, building a map, and then planning a route as many traditional autonomous driving systems do. The README describes two intended use cases. The first is for fully driverless robotaxi applications, where the camera-based model is combined with physical sensors such as LIDAR (which uses laser pulses to measure distances) or RADAR (which uses radio waves). The second is for what the industry calls L2++ driver assistance, meaning hands-free point-to-point navigation where a human remains available to take over but the car handles most of the driving. To get started, the README points to a requirements file for installing dependencies and to a Model folder inside the repository for training and running the model. No further setup detail is given in the README itself. The project uses PyTorch, a widely used machine learning framework, and is licensed under the Apache 2.0 open-source license. It is a very new repository with only 14 stars at the time of this writing and a short README that does not yet document the full scope of the system.

Copy-paste prompts

Prompt 1

I cloned auto_fsd from autowarefoundation. How do I install the dependencies from the requirements file and run the model on a sample camera video?

Prompt 2

I want to use auto_fsd for L2++ driver assistance rather than full robotaxi mode. What model configuration changes are needed and what does the training script in the Model folder expect as input?

Prompt 3

How does auto_fsd's end-to-end approach differ from traditional autonomous driving stacks like Autoware that separate object detection, mapping, and route planning?

Prompt 4

I want to fine-tune the auto_fsd PyTorch model on my own dashcam dataset. What input format does the training script expect and what GPU requirements are recommended?

Open on GitHub → Explain another repo

← autowarefoundation on gitmyhub — every repo by this author, as a profile.

Verify against the repo before relying on details.