This repository contains machine learning components and tools to advance sunnypilot's capabilities in autonomous driving.
The repository includes implementations and tools for:
- Vision Encoders: Neural network components for processing visual data from cameras.
- Vision Classifiers: Models for classifying objects and scenes in driving environments, such as street lights, pedestrians, etc...
- Machine Learning Tests: Test suites for validating ML models and modeld codebase.
- Testing Capabilities: Validated integration testing for troubleshooting and analyzing model performance.
- Driving Model Merge Scripts: Tools to combine multiple driving models into one.
- Quantizers: Algorithms for model quantization to optimize performance and size, able to downsample 1 GB models to 200MiB models.
- Model-Specific Code: Offloaded code from the sunnypilot repository to help better the dev experience.
To set up the development environment, run the appropriate setup script for your operating system:
- macOS:
./tools/setup/mac_setup.sh - Ubuntu:
./tools/setup/ubuntu_setup.sh
After running the script, activate the virtual environment in your shell:
source .venv/bin/activate
The params module uses a compiled C++ library for default values. Build it with:
scons
-
navigation: module containing mapbox navigation features created to be transitioned to sunnypilot as a semi-offline navigation naemon. This module will map turn desires, keep left/right, and eventually auto lane changes to navigation instructions to provide a more comfortable user experience. -
tools: setup scripts and test runners.
We welcome both pull requests and issues on GitHub. Bug fixes are encouraged.
Pull requests should be against the most current master branch.
sunnypilot autonomy is released under the MIT License.
THIS IS ALPHA QUALITY SOFTWARE FOR RESEARCH PURPOSES ONLY. THIS IS NOT A PRODUCT. YOU ARE RESPONSIBLE FOR COMPLYING WITH LOCAL LAWS AND REGULATIONS. NO WARRANTY EXPRESSED OR IMPLIED.
For full license terms, please see the LICENSE file.