# GreenWAM preprint reports shared prediction and control across robot families

_Published Wednesday, October 14, 2026 at 1:52 AM EDT · Robotics, AI, Science · Latest · Tier 2 — Notable_

Researchers report that GreenWAM can both predict how robot actions change a scene and choose actions across different robot bodies. Their preprint describes a shared set of joint coordinates for 20 robot families, joint training for prediction and control, and an action component separated from the generator.

In tests on a real humanoid, GreenWAM succeeded more often than policies adapted from π₀.₅ and Cosmos 3 Edge. The researchers also distilled it into a single-step policy for real-time control. The reported comparisons cover selected benchmarks and a real humanoid adaptation.

## Sources

- [arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30](https://arxiv.org/abs/2610.18774v1)

---
Canonical: https://techandbusiness.org/newswire/4a6RNLzaQuyyMPXLwhVWOr
Published: 2026-10-14T05:52:41.406Z
Story chronology: 2026-10-13T17:31:29.000Z
Retrieved: 2026-10-14T09:01:11.941Z
Publisher: Tech & Business (techandbusiness.org)