Liquid's d1-3B decision model drives robot skills at 157ms on Intel Arc B580
helloiamleonie · x · 2026-10-19
A demo shows Liquid AI's d1-3B multimodal decision model picking robot skills like walk, turn, pick, and sit, with a trained policy executing each move — running in just 157ms on an Intel Arc B580. The author also compiled a thread of other use cases for the d1 decision model family, showing small decision models can drive robots in real time on consumer hardware.
More from Embodied
- Humanoid robots gain a sense of touch: the rise of feeling robots — TinfoilTricorn · 2026-10-19
- ARC-SORT: A Training-Free Real-Time Tracker Cuts Identity Switches in Sports Broadcasts — Girish Kumar Adari · 2026-10-19
- Shipbuilder HII's $900M robotics deals point to engineering firms selling parts, not software — burhop · 2026-10-19
- d1-omni-600M runs solar-powered on a Raspberry Pi 5, with a sub-$125 productization plan — helloiamleonie · 2026-10-19
- AUAR ships micro-factories in shipping containers: 200 units could clear UK's annual housing backlog — sarahdrinkwater · 2026-10-19
- Zhejiang University survey frames smart glasses as first-person intelligence platforms with L0-L5 scale — 机器之心 · 2026-10-19