Synthesizing Physically Plausible Human Motions
in 3D Scenes


Liang Pan1   Jingbo Wang2   Buzhen Huang1   Junyu Zhang1
Haofan Wang3   Xu Tang3   Yangang Wang1

Southeast University1 Shanghai AI Lab2 Xiaohongshu Inc.3

International Conference on 3D Vision (3DV 2024)

[Paper]   [Code]


We present a physics-based character control framework for synthesizing human-scene interactions. Recent advances adopt physics simulation to mitigate artifacts produced by data-driven kinematic approaches. However, existing physics-based methods mainly focus on single-object environments, resulting in limited applicability in realistic 3D scenes with multi-objects. To address such challenges, we propose a framework that enables physically simulated characters to perform long-term interaction tasks in diverse, cluttered, and unseen 3D scenes. The key idea is to decouple human-scene interactions into two fundamental processes, Interacting and Navigating, which motivates us to construct two reusable Controllers, namely InterCon and NavCon. Specifically, InterCon uses two complementary policies to enable characters to enter or leave the interacting state with a particular object (e.g., sitting on a chair or getting up). To realize navigation in cluttered environments, we introduce NavCon, where a trajectory following policy enables characters to track pre-planned collision-free paths. Benefiting from the divide and conquer strategy, we can train all policies in simple environments and directly apply them in complex multi-object scenes through coordination from a rule-based scheduler.

Poster


Motivation

Previous works [1, 2]

Ours

Existing physics-based scene interaction approaches cannot generalize to multi-object scenes due to the lack of two crucial abilities: (1) continuous interaction and (2) obstacle avoidance.

Pipeline


Given a multi-object 3D scene, our goal is to synthesize long-term motion sequences by controlling a physics-based character to perform a series of scene interaction tasks. First, our system employs an interaction controller to provide two primary actions, i.e., sitting down and getting up. Second, we introduce a navigation controller to acquire another action, i.e., collision-free trajectory following. Finally, a rule-based action scheduler is exploited to obtain outputs by organizing reusable low-level actions according to user-designed instructions.

Generated Long-term Motions in Diverse 3D Scenes


Images

Videos

Extensibility
By training an additional interaction controller, a new skill of lying down can be seamlessly integrated into our system, which demonstrates the strong extensibility of our approach. Given two interaction controllers, our extended system enables the physics-based character to first lie on the sofa, then sits on the chair, and finally lie on the bed, exhibiting more diverse long-term interactions.

Citation

@inproceedings{pan2024synthesizing,
    title={Synthesizing physically plausible human motions in 3d scenes},
    author={Pan, Liang and Wang, Jingbo and Huang, Buzhen and Zhang, Junyu and Wang, Haofan and Tang, Xu and Wang, Yangang},
    booktitle={2024 International Conference on 3D Vision (3DV)},
    pages={1498--1507},
    year={2024},
    organization={IEEE}
}

References