Morph: A Motion-free Physics Optimization Framework for Human Motion Generation

Abstract

Human motion generation has been widely studied due to its crucial role in areas such as digital humans and humanoid robot control. However, many current motion generation approaches disregard physics constraints, frequently resulting in physically implausible motions with pronounced artifacts such as floating and foot sliding. Meanwhile, training an effective motion physics optimizer with noisy motion data remains largely unexplored. In this paper, we propose \textbf{Morph}, a \textbf{Mo}tion-F\textbf{r}ee \textbf{ph}ysics optimization framework, consisting of a Motion Generator and a Motion Physics Refinement module, for enhancing physical plausibility without relying on expensive real-world motion data. Specifically, the motion generator is responsible for providing large-scale synthetic, noisy motion data, while the motion physics refinement module utilizes these synthetic data to learn a motion imitator within a physics simulator, enforcing physical constraints to project the noisy motions into a physically-plausible space. Additionally, we introduce a prior reward module to enhance the stability of the physics optimization process and generate smoother and more stable motions. These physically refined motions are then used to fine-tune the motion generator, further enhancing its capability. This collaborative training paradigm enables mutual enhancement between the motion generator and the motion physics refinement module, significantly improving practicality and robustness in real-world applications. Experiments on both text-to-motion and music-to-dance generation tasks demonstrate that our framework achieves state-of-the-art motion quality while improving physical plausibility drastically. Project page: https://interestingzhuo.github.io/Morph-Page/.

Figures (7)

• Figure 1: Examples of physical inconsistencies in generations.
• Figure 2: An overview of the Morph framework. Morph comprises a Motion Generator and a Motion Physics Refinement module. Morph employs a two-stage training process: Motion Physics Refinement module training and Motion Generator fine-tuning. And a Imitation Selection Operation is employed to ensure the motion quality after physics refinement. The solid curved arrows on the left and right (in orange and green) represent the iterative, collaborative optimization between Stage 1 and Stage 2.
• Figure 3: Qualitative comparison between our Morph-MoMask and MoMask in text-to-motion task. Morph-MoMask significantly reduces physical artifacts such as leaning forward, floating and penetration.
• Figure 4: A flowchart illustrating the data preprocessing process. The parameters are calculated from the first frame and then applied to all generated motion sequences before they are fed into the MPR module.
• Figure 5: T-sne of motion and text distribution between MG and Morph.