PoseShield: Neural Collision Fields for Human Self-Collision Resolution

2026-06-28

AI Summary

This work addresses self-collision artifacts in SMPL-based human pose estimation and motion generation caused by extreme joint configurations or randomly synthesized poses. The authors propose a constrained optimization method that models self-collision correction directly in a low-dimensional pose space. Their approach introduces, for the first time, a theory-driven neural implicit collision constraint that operates without mesh-space manipulations or heuristic penalties, and incorporates Eikonal equation regularization to ensure numerical stability. The resulting module functions as a generator-agnostic post-processing component, generalizing seamlessly to arbitrary motion sequences. Evaluated on a newly constructed SMPL pose benchmark, the method achieves a collision-resolution success rate of 95.8%, substantially outperforming current state-of-the-art techniques.

Abstract

Self-collision remains a persistent challenge in SMPL-based human pose estimation and motion generation. Under extreme articulations or stochastic motion synthesis, generated meshes frequently exhibit self-penetrations, leading to physically implausible results. We propose PoseShield, a neural collision constraint defined directly in SMPL pose space. We formulate collision correction as a constrained optimization problem and connect the learned constraint with the Eikonal equation. Enforcing Eikonal regularization ensures non-vanishing gradients near the collision boundary, improving numerical stability and robustness of the optimization process. Unlike prior methods that operate in the mesh space or rely on heuristic penalties, our approach operates directly in the low-dimensional space of human poses and is theoretically grounded. The same learned constraint extends to human motion sequences, providing a generator-agnostic post-hoc collision corrector without retraining the underlying motion model. Experiments on a newly constructed SMPL pose benchmark show that our method achieves a 95.8% success rate and outperforms state-of-the-art baselines.

Problem

• self-collision
• SMPL
• human pose estimation
• motion generation
• mesh penetration

Innovation

• Neural Collision Fields
• SMPL Pose Space
• Eikonal Regularization
• Self-Collision Resolution
• Generator-Agnostic Correction