Tsinghua and Infinigence open-source APXInf, cutting embodied model latency 10.7x on Thor
机器之心 · wechat · 2026-09-29
Tsinghua University, Infinigence (无问芯穹) and Shanghai Jiao Tong University open-sourced APXInf, an on-device inference engine for embodied models. Without modifying π0.5, full-stack optimization cuts inference latency from 278ms to 26ms on Jetson Thor (FP8) — a 10.7x speedup reaching 38.46Hz real-time control — while LIBERO-10 success rate stays at 92.2%, matching the reference implementation. On Orin, optimization dropped latency from 1300ms to 119ms (10.9x). APXInf uses model-family-specific execution paths, CUDA Graph capture, pre-allocated memory and a Rust-based runtime, and turns model onboarding/tuning into an agent-executable workflow so humans only set standards and acceptance criteria. Already adapted for π0-fast, GR00T and QwenDrive, with AMD and domestic chip backends planned. The tech inherits from Mizar, preinstalled on 10M+ Lenovo AI PCs.
More from Embodied
- Leaked OpenAI 'dot' details show raising phone to ear triggers ChatGPT Voice — koltregaskes · 2026-09-29
- Uniformation S10 Robotic Resin Printer Handles Printing, Washing and Drying Automatically — philfung · 2026-09-29
- REALM generates reactive listener facial motion, deployed on an Ameca humanoid robot — MacquarieUni · 2026-09-29
- Korea's Diden Spider: a magnetic-legged welding robot that climbs shipyard walls — MickeySteamboat · 2026-09-29
- "Terminator" robot demoed at REK arena as man-vs-machine extreme sport takes shape — cixliv · 2026-09-29
- AIgirl demo shows a robot making dinner — miamiamorph · 2026-09-29