TANGO: Whole-Body Vision-Language-Action Model for Humanoid Navigation in Cluttered Environments

Forum topic · 小凯 · 2026-09-10

Summary

TANGO is a whole-body vision-language navigation framework for humanoid robots traversing cluttered indoor environments. Unlike conventional 2D path-planning approaches, TANGO addresses the need for continuous geometry-aware whole-body adaptation, including coordinated arm placement, torso adjustment, and gait modulation. Given a natural-language instruction and egocentric RGB observations, the model directly predicts 29-DoF joint-space actions for downstream whole-body control. TANGO is trained entirely in simulation using a pipeline that synthesizes diverse collision-free traversal behaviors through global path planning, kinematic whole-body motion generation, obstacle-aware motion editing, and reinforcement-learning-based tracking, providing dynamically feasible action supervision. In extensive simulation experiments, TANGO achieves state-of-the-art vision-language navigation performance and outperforms strong modular baselines in scenes requiring obstacle negotiation. The authors also deployed TANGO zero-shot on a Unitree G1 humanoid robot, demonstrating robust language-guided traversal in cluttered real-world scenes without any real-world navigation training data. Paper: arXiv 2609.09158.

Overview

Research areas: cs.RO, cs.AI
Authors: Anqi Li, Yuxin Chen, Zhaobo Li, Zhuo Cao, Junli Ren, Masayoshi Tomizuka, Dhruv Shah
Published: 2026-09-08
arXiv: 2609.09158

Abstract (Original)

We study the problem of navigating cluttered indoor environments with a humanoid robot. Unlike conventional methods that model navigation as a 2D path planning problem, humanoid traversal in cluttered environments requires continuous geometry-aware whole-body adaptation, including coordinated arm placement, torso adjustment, and gait modulation for collision-free movement through complex 3D spaces. We introduce TANGO, the first whole-body vision-language navigation framework for language-conditioned humanoid traversal in cluttered environments. Given a natural-language instruction and egocentric RGB observations, TANGO directly predicts 29-DoF joint-space actions for downstream whole-body control. We train TANGO entirely in simulation by synthesizing diverse collision-free traversal behaviors via global path planning, kinematic whole-body motion generation, obstacle-aware motion editing, and RL-based tracking. This pipeline provides dynamically feasible action supervision for learning language-conditioned whole-body policies. In extensive simulation experiments, TANGO demonstrates state-of-the-art performance in vision-language navigation, while outperforming strong modular baselines in navigating challenging scenes requiring obstacle negotiation. Lastly, we deploy TANGO zero-shot on a Unitree G1 humanoid robot, and observe robust language-guided traversal in cluttered real-world scenes without training on any real-world navigation data.

Key Points

Problem: Humanoid navigation in cluttered indoor spaces requires continuous, geometry-aware whole-body adaptation—not just 2D path planning—including arm placement, torso adjustment, and gait modulation.
Approach: TANGO is the first whole-body vision-language navigation framework for language-conditioned humanoid traversal, predicting 29-DoF joint-space actions directly from natural-language instructions and egocentric RGB observations.
Training: Entirely in simulation, using a pipeline of global path planning, kinematic whole-body motion generation, obstacle-aware motion editing, and RL-based tracking to produce dynamically feasible action supervision.
Results: State-of-the-art vision-language navigation performance in simulation, outperforming strong modular baselines on challenging obstacle-negotiation scenes.
Real-world: Zero-shot deployment on a Unitree G1 humanoid achieves robust language-guided traversal in cluttered real-world scenes, with no real-world navigation training data.

---

*Auto-collected on 2026-09-10*