XPENG's AUGUstus VLA model hits SOTA on NAVSIM and deploys on real vehicles at 12Hz
XPENG-AI · hf · 2026-10-14
XPENG-AI, with Peking and Fudan universities, releases AUGUstus, an autonomous-driving VLA model unifying diffusion-based action generation with autoregressive CoT understanding, where understanding only serves generation.
- AUGUstus-mode generates actions before CoT so inference needs only action output, enabling low latency
- AUGUstus-GRPO jointly optimizes actions and CoT
- SOTA planning without scorer: 94.6 PDMS on NAVSIMv1, 90.7 EPDMS on NAVSIMv2; nuScenes L2 0.27m, collision rate 0.05%
- Trained on 512 H800 GPUs for 2 days, validated on 1,000 km internal data, preliminarily deployed on a real vehicle at 12Hz
More from Embodied
- NAVER's WATCH: Cross-Video Human-Video Pretraining Lifts Robot Success Rate to 77.7% — naver-ai · 2026-10-14
- Robot Dog Tries to Climb Its Own Back: Conitzer's Latest Funny AI Fail — conitzer · 2026-10-14
- This Week in New Models: RoboJEPA, Long-WAM, VioLA and 4 More to Know — TheTuringPost · 2026-10-14
- EMHO: Self-Evolving Agent Harness Boosts Frozen Embodied Models via Experience Traces — Hyun Jung Lee · 2026-10-14
- Robotics startup Origin teases 'major announcement' with robot arm image — vaibhavbetter · 2026-10-14
- Paradigm appears to be getting into robotics with LeRobot arm kits — tensorqt · 2026-10-14