# HumanVerse dataset and λ0 model released for humanoid loco-manipulation

> A new 500‑hour egocentric human dataset and a foundation model achieve up to 90% success on whole‑body robot tasks.

Oossa · 2026-10-15 · https://oossa.com/en/humanverse-dataset-and-0-model-released-for-humanoid-loco-manipulation

Researchers have published HumanVerse, a 500‑hour collection of egocentric video paired with full‑body and hand motion. They also released λ0, a humanoid foundation model trained on this data. λ0 was tested on the Unitree G1 robot and reached 90.0% success on the SIMPLE benchmark and 62.5% on real‑world loco‑manipulation tasks, all without robot‑specific demonstrations.

## The facts

- HumanVerse contains 500 hours of open‑world human loco‑manipulation data.
- λ0 achieves 90.0% success on the SIMPLE benchmark.

## Why it matters

The dataset and model let robot builders teach whole‑body manipulation to humanoid robots without collecting costly robot demonstrations.

## Sources & references

1. [$\lambda_0$: Scaling Open-World Egocentric Human Data for Humanoid Loco-Manipulation](https://arxiv.org/abs/2610.18950) – arXiv cs.RO, 2026-10-15

Last updated: 2026-10-15