HumanVLA: Towards Vision-Language Directed Object Rearrangement by Physical Humanoid

TL;DR

This work introduces HumanVLA for general object rearrangement directed by practical vision and language and introduces a novel Human-in-the-Room dataset encompassing various rearrangement tasks.

Abstract

Physical Human-Scene Interaction (HSI) plays a crucial role in numerous applications. However, existing HSI techniques are limited to specific object dynamics and privileged information, which prevents the development of more comprehensive applications. To address this limitation, we introduce HumanVLA for general object rearrangement directed by practical vision and language. A teacher-student framework is utilized to develop HumanVLA. A state-based teacher policy is trained first using goal-conditioned reinforcement learning and adversarial motion prior. Then, it is distilled into a vision-language-action model via behavior cloning. We propose several key insights to facilitate the large-scale learning process. To support general object rearrangement by physical humanoid, we introduce a novel Human-in-the-Room dataset encompassing various rearrangement tasks. Through extensive experiments and analysis, we demonstrate the effectiveness of the proposed approach.

Figures (12)

• Figure 1: HumanVLA performs various object rearrangement tasks directed by the egocentric vision and natural language instructions.
• Figure 2: An overview of learning state-based HumanVLA-Teacher policy using goal-conditioned reinforcement learning and adversarial motion prior.
• Figure 3: Left: An overview of learning HumanVLA by mimicking teacher action and active rendering action. Right: Comparison between w/ and w/o active rendering. Active rendering leads to a more informative perception of human-object relationships.
• Figure 4: Qualitative results. The color transitions from green to yellow as the task progresses.
• Figure 5: The task generation process of HITR dataset.