Xiaomi's DriveZero: end-to-end driving that beats humans without trajectory supervision
abursuc · x · 2026-09-10
Xiaomi's L3 team unveiled DriveZero, an end-to-end visual driving policy released as a tech report on arXiv.
- Learns without human trajectory supervision yet surpasses human drivers on NAVSIM
- Sets SOTA on both NAVSIMv2 and HUGSIM benchmarks
- Includes real-world deployment demonstration
- Builds on open models like RADIO, SAM, DINOv3, and Depth Anything
Code and checkpoints are not yet released; the GitHub repo (XiaomiAutoL3/DriveZero) currently hosts the tech report, project page, and a TODO list.
Related event: Xiaomi L3 Team Open-Sources DriveZero End-to-End Driving System(2 posts)→
More from Embodied
- WanGP v18.81 adds Tencent's Hy-Nomas world model, playable on 8GB VRAM — cocktailpeanut · 2026-10-27
- MS-Human-700: open-source human model with 700 muscle actuators lands in MuJoCo — yisongyue · 2026-10-27
- NVIDIA open-sources FLORA, a robotics data engine ~125x cheaper than GPT-6 Astra — nvidia · 2026-10-27
- Embodied AI startups need a different operational discipline, investor says — ___Mufasaa · 2026-10-27
- π0.5 study: raising demos from 10 to 438 lifts Place Ball in Bowl success from 7% to 83% — stingning · 2026-10-27
- GPT-6 Astra controls dexterous hands zero-shot at just 5.2% success in 500 trials — stingning · 2026-10-27