Comma.ai Recaps Hilarious Early Bugs in Self-Driving Models
dosco · x · 2026-08-03
comma.ai published a blog post detailing entertaining bugs from the early days of their openpilot self-driving system and their root causes. The author notes that most driving improvements came from fixing blatant engineering bugs rather than implementing flashy new ML techniques.
Key cases include:
- Rightward Drift (2017): Early systems lacked closed-loop testing. The model's trajectory predictions relied on a localizer using gyro measurements. While static gyro bias was calibrated at startup, device heating once driving caused the bias to change. Since the device never rested, the bias wasn't recalibrated, skewing all training data and causing the car to drift right. This was fixed with a more comprehensive localizer.
More from Embodied
- CS-JEPA: Decentralized Predictive Architecture for Swarm Robotics — ITMO · 2026-08-03
- ODEWorld: A Continuous-Time Latent World Model via ODE — Dongxiu Liu · 2026-08-03
- N_0-TWAM: First Large-Scale Tactile-Native World-Action Model for Robots — NeoteAIEmbodied · 2026-08-03
- Interview with Luo Ping: The Core of Embodied AI Models is Bridging the Human-to-Robot Data Loop — 机器之心 · 2026-08-03
- HUST & Huawei's TurboVLA Bypasses LLM for 32Hz Real-Time Robot Control — 机器之心 · 2026-08-03
- VOCA: Boosting Visual Odometry Performance Using Codec Information — rsasaki0109 · 2026-08-03