PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation

• URL: http://arxiv.org/abs/2605.14269v1
• Date: Thu, 14 May 2026 02:12:13 GMT
• Title: PhyMotion: Structured 3D Motion Reward for Physics-Grounded Human Video Generation
• Abstract summary: Existing video rewards primarily rely on 2D perceptual signals, without explicitly modeling the 3D body state, contact, and dynamics underlying articulated human motion.<n>We propose PhyMotion, a structured, fine-grained motion reward that grounds recovered 3D human trajectories in a physics simulator and evaluates motion quality along multiple dimensions of physical feasibility.<n>Experiments show that PhyMotion achieves stronger correlation with human judgments than existing reward formulations.
• Score: 75.96085587438279
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Generating realistic human motion is a central yet unsolved challenge in video generation. While reinforcement learning (RL)-based post-training has driven recent gains in general video quality, extending it to human motion remains bottlenecked by a reward signal that cannot reliably score motion realism. Existing video rewards primarily rely on 2D perceptual signals, without explicitly modeling the 3D body state, contact, and dynamics underlying articulated human motion, and often assign high scores to videos with floating bodies or physically implausible movements. To address this, we propose PhyMotion, a structured, fine-grained motion reward that grounds recovered 3D human trajectories in a physics simulator and evaluates motion quality along multiple dimensions of physical feasibility. Concretely, we recover SMPL body meshes from generated videos, retarget them onto a humanoid in the MuJoCo physics simulator, and evaluate the resulting motion along three axes: kinematic plausibility, contact and balance consistency, and dynamic feasibility. Each component provides a continuous and interpretable signal tied to a specific aspect of motion quality, allowing the reward to capture which aspects of motion are physically correct or violated. Experiments show that PhyMotion achieves stronger correlation with human judgments than existing reward formulations. These gains carry over to RL-based post-training, where optimizing PhyMotion leads to larger and more consistent improvements than optimizing existing rewards, improving motion realism across both autoregressive and bidirectional video generators under both automatic metrics and blind human evaluation (+68 Elo gain). Ablations show that the three axes provide complementary supervision signals, while the reward preserves overall video generation quality with only modest training overhead.

Related papers

• AnyLift: Scaling Motion Reconstruction from Internet Videos via 2D Diffusion [41.002383749585924]
  We introduce a two-stage framework leveraging 2D diffusion that reconstructs 3D human motion and HOI from Internet videos.<n>In the first stage, we synthesize multi-view 2D motion data for each domain, leveraging 2D keypoints extracted from Internet videos.<n>In the second stage, a camera-conditioned multi-view 2D motion diffusion model is trained on the domain-specific synthetic data to recover 3D human motion and 3D HOI in the world space.
  arXiv  Detail & Related papers  (2026-04-20T05:15:38Z)
• Learning to Control Physically-simulated 3D Characters via Generating and Mimicking 2D Motions [23.080971732537886]
  Mimic2DM is a novel motion imitation framework that learns the control policy directly from 2D keypoint trajectories extracted from videos.<n>We show that the proposed approach is versatile and can effectively learn to synthesize physically plausible and diverse motions across a range of domains.
  arXiv  Detail & Related papers  (2025-12-09T11:30:56Z)
• PhysHMR: Learning Humanoid Control Policies from Vision for Physically Plausible Human Motion Reconstruction [52.44375492811009]
  We present PhysHMR, a unified framework that learns a visual-to-action policy for humanoid control in a physics-based simulator.<n>A key component of our approach is the pixel-as-ray strategy, which lifts 2D keypoints into 3D spatial rays and transforms them into global space.<n>PhysHMR produces high-fidelity, physically plausible motion across diverse scenarios, outperforming prior approaches in both visual accuracy and physical realism.
  arXiv  Detail & Related papers  (2025-10-02T21:01:11Z)
• A Plug-and-Play Physical Motion Restoration Approach for In-the-Wild High-Difficulty Motions [56.709280823844374]
  We introduce a mask-based motion correction module (MCM) that leverages motion context and video mask to repair flawed motions.<n>We also propose a physics-based motion transfer module (PTM), which employs a pretrain and adapt approach for motion imitation.<n>Our approach is designed as a plug-and-play module to physically refine the video motion capture results, including high-difficulty in-the-wild motions.
  arXiv  Detail & Related papers  (2024-12-23T08:26:00Z)
• MotionBERT: A Unified Perspective on Learning Human Motion Representations [46.67364057245364]
  We present a unified perspective on tackling various human-centric video tasks by learning human motion representations from large-scale and heterogeneous data resources. We propose a pretraining stage in which a motion encoder is trained to recover the underlying 3D motion from noisy partial 2D observations. We implement motion encoder with a Dual-stream Spatio-temporal Transformer (DSTformer) neural network.
  arXiv  Detail & Related papers  (2022-10-12T19:46:25Z)
• Skeleton2Humanoid: Animating Simulated Characters for Physically-plausible Motion In-betweening [59.88594294676711]
  Modern deep learning based motion synthesis approaches barely consider the physical plausibility of synthesized motions. We propose a system Skeleton2Humanoid'' which performs physics-oriented motion correction at test time. Experiments on the challenging LaFAN1 dataset show our system can outperform prior methods significantly in terms of both physical plausibility and accuracy.
  arXiv  Detail & Related papers  (2022-10-09T16:15:34Z)
• D&D: Learning Human Dynamics from Dynamic Camera [55.60512353465175]
  We present D&D (Learning Human Dynamics from Dynamic Camera), which leverages the laws of physics to reconstruct 3D human motion from the in-the-wild videos with a moving camera. Our approach is entirely neural-based and runs without offline optimization or simulation in physics engines.
  arXiv  Detail & Related papers  (2022-09-19T06:51:02Z)
• Differentiable Dynamics for Articulated 3d Human Motion Reconstruction [29.683633237503116]
  We introduce DiffPhy, a differentiable physics-based model for articulated 3d human motion reconstruction from video. We validate the model by demonstrating that it can accurately reconstruct physically plausible 3d human motion from monocular video.
  arXiv  Detail & Related papers  (2022-05-24T17:58:37Z)
• Contact and Human Dynamics from Monocular Video [73.47466545178396]
  Existing deep models predict 2D and 3D kinematic poses from video that are approximately accurate, but contain visible errors. We present a physics-based method for inferring 3D human motion from video sequences that takes initial 2D and 3D pose estimates as input.
  arXiv  Detail & Related papers  (2020-07-22T21:09:11Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.