# Zero-HLM: Generalizable Humanoid Loco-Manipulation without Real Robot Data

Preprint · Dianyi Yang, Mengya Liu, Zhumei Wang, Yu Liu, Xiaochen Zhang, Yixuan Li, Le Ma, Yuhang Yang, Weiqi Huang, Wei Liang, Baoxiong Jia, Siyuan Huang

- Archive id: arxiv:2610.35075
- Published: 2026-10-26
- arXiv: https://arxiv.org/abs/2610.35075
- Citations: 0
- Source URL: https://arxiv.org/abs/2610.35075
- PDF: https://arxiv.org/pdf/2610.35075
- Provenance: arxiv

## Abstract

Learning generalizable humanoid loco-manipulation remains bottlenecked by costly real-world demonstrations. Simulation provides a scalable alternative, but requires visually faithful, physics-ready scenes and executable whole-body motions. We present Zero-HLM, a real-to-sim-to-real framework that transforms smartphone captures and simulated source demonstrations into scalable training data for humanoid loco-manipulation. Gaussian-guided agentic real-to-sim reconstruction aligns visual and physical scenes; motion-environment co-adaptation uses rollout feedback to correct whole-body references and interaction geometry. We generate 4,800 physically validated demonstrations for zero-shot transfer across three Unitree G1 tasks. Simulation-trained GR00T N1.7 achieves 66.7% mean full-task success, versus 46.7% with real-data training. On generalized configurations, expanded simulation training raises pooled success from 23% for base-trained real-data policies to 69%, without additional real demonstrations. Project page: https://young-bit.github.io/zero-hlm/

_Abstract reproduced under CC0-1.0 (arXiv descriptive metadata)._

Topics: [Robot Learning & Embodied AI](https://space-kitty.com/topics/robotics-embodied)

---
Space Kitty Archive · https://space-kitty.com · API: https://space-kitty.com/api/v1 · MCP: https://space-kitty.com/mcp · Premium: $0.20/request via x402 (Base Sepolia (testnet)). Metadata: arXiv (CC0), OpenAlex (CC0), Europe PMC (CC BY/CC0 works), ClinicalTrials.gov.