# Preprint reports keyframe control gains for Unitree humanoid

_Published Wednesday, September 16, 2026 at 11:07 PM EDT · Robotics, Science · Latest · Tier 2 — Notable_

A preprint describes KINO, a control framework that uses motion keyframes between vision-language planning and reinforcement-learning control for humanoid loco-manipulation. The authors report that a saliency-based keyframe training method raised end-to-end task success from 44% to 92% when sparse VLM keyframes were used. They evaluated pickup, transport and placement tasks in simulation and on a Unitree G1 humanoid, including one- and two-handed manipulation.

## Sources

- [arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30](https://arxiv.org/abs/2609.18869v1)

---
Canonical: https://techandbusiness.org/newswire/tqTJeVt-ODkwjzmE3S6sBn
Published: 2026-09-17T03:07:13.450Z
Story chronology: 2026-09-16T16:07:33.000Z
Retrieved: 2026-09-17T05:31:37.194Z
Publisher: Tech & Business (techandbusiness.org)