VIRAL: Visual Sim-to-Real at Scale for Humanoid Loco-Manipulation

TL;DR

VIRAL is a scalable visual sim-to-real framework enabling humanoid robots to learn loco-manipulation skills entirely in simulation and deploy them zero-shot in real-world scenarios, achieving expert-level performance.

Contribution

The paper introduces VIRAL, a novel large-scale visual sim-to-real approach with a teacher-student design, enabling zero-shot transfer of humanoid loco-manipulation skills.

Findings

Scaling simulation to 64 GPUs improves training reliability.

VIRAL achieves up to 54 continuous loco-manipulation cycles in real-world deployment.

RGB-based policy generalizes across diverse environments without fine-tuning.

Abstract

A key barrier to the real-world deployment of humanoid robots is the lack of autonomous loco-manipulation skills. We introduce VIRAL, a visual sim-to-real framework that learns humanoid loco-manipulation entirely in simulation and deploys it zero-shot to real hardware. VIRAL follows a teacher-student design: a privileged RL teacher, operating on full state, learns long-horizon loco-manipulation using a delta action space and reference state initialization. A vision-based student policy is then distilled from the teacher via large-scale simulation with tiled rendering, trained with a mixture of online DAgger and behavior cloning. We find that compute scale is critical: scaling simulation to tens of GPUs (up to 64) makes both teacher and student training reliable, while low-compute regimes often fail. To bridge the sim-to-real gap, VIRAL combines large-scale visual domain randomization over lighting, materials, camera parameters, image quality, and sensor delays--with real-to-sim alignment of the dexterous hands and cameras. Deployed on a Unitree G1 humanoid, the resulting RGB-based policy performs continuous loco-manipulation for up to 54 cycles, generalizing to diverse spatial and appearance variations without any real-world fine-tuning, and approaching expert-level teleoperation performance. Extensive ablations dissect the key design choices required to make RGB-based humanoid loco-manipulation work in practice.