Open Source Project: Distill openpilot to Train Custom Autonomous Driving Models
yassineyousfi_ · x · 2026-09-01
A new open-source project, openpilot.distill, enables developers to train state-of-the-art autonomous driving models using distilled knowledge from openpilot.
Key Features and Workflow:
- Localization & Distillation: Users can either use a comma four device or download the comma1M dataset from Hugging Face. The process involves running a localization pipeline and then training a smaller student model using targets distilled from the openpilot model.
- Simulation Fine-tuning: Supports DAgger-style post-training within a learned world model simulator to refine driving policies.
- Custom Driving Styles: Developers can upload their own drive segments (video and rlogs) to train models that mimic their specific driving style.
The repository provides the full codebase and step-by-step instructions, making it accessible even without dedicated proprietary hardware.
More from Embodied
- Opus-assisted open design: a $31k low-field head MRI sketched in hours — DreamFly_13 · 2026-10-18
- DIY robotics build: servo mechanism tested, wide yaw range but limited pitch — _Stocko_ · 2026-10-18
- Brain implants are restoring speech to people with ALS, Science reports — EricMTrautmann · 2026-10-18
- Meta wearables VP runs his AI agent across 3 phones and 23 pairs of glasses — nikola_mr64990 · 2026-10-18
- AMI-EV: spinning prism gives event cameras stable texture, Science Robotics 2024 — RexDouglass · 2026-10-18
- Neuralink patients control robotic arms, drones and games with thought alone — XFreeze · 2026-10-18