HSImul3R (ECCV 2026): simulation-ready human-scene reconstruction judged by physics, not looks
jiqizhixin · x · 2026-09-28
Xiaoxiao Robot, with NTU S-Lab and Shanghai AI Laboratory, present HSImul3R, accepted at ECCV 2026, targeting the long-standing "perception-to-simulation gap": reconstructions that look correct often collapse or clip through scenes in physics simulators.
- Key shift: instead of judging visual quality or alignment, the framework makes gravity stability, real contact, and interaction stability the core criteria
- Claimed as the first framework for stable, simulation-ready human-scene interaction reconstruction from uncalibrated sparse-view input, with monocular video support
- The team also builds HSIBench, a benchmark testing whether reconstructions can stably interact in physical simulation
More from Embodied
- Sharpa Robotics wows IROS by catching randomly falling foam cylinders — chris_j_paxton · 2026-09-28
- Airbound's drone carries more than it weighs, taking on Amazon Prime Air — Not Boring (Packy McCormick) · 2026-09-28
- World's largest humanoid robot livestream performance draws crowd of 10,000+ — kernelangus420 · 2026-09-28
- One-arm VR intervention during bimanual DAgger feels like an AI brain chip — neurosp1ke · 2026-09-28
- USTC's VLA-Precision: online RL for VLAs hits 98.3% success on precision chemistry tasks — ustc · 2026-09-28
- Robot training fix: cutting noise injection from 2.5% to 0.25% of joint range removes jerkiness — DominiqueCAPaul · 2026-09-28