ActiveAlign uses motion and touch to close Real2Sim2Real gaps
Original title: ActiveAlign: Aligning Egocentric Manipulation via Active Perception
On a Unitree G1, the method supports button presses and slider moves at millimeter scale.
ActiveAlign is a framework that uses active perception to reduce Real-to-Sim-to-Real gaps caused by passive observation. ActiveAlign-Waving calibrates hand-eye geometry from hand motion and differentiable rendering. ActiveAlign-Touching refines object alignment with tactile feedback to millimeter-level precision. The authors say waving can build simulation environments from open-sourced uncalibrated manipulation datasets, and they demonstrate fine manipulation on a real Unitree G1.
- Passive perception is described as the source of real-to-sim observation gaps.
- Waving exploits active hand motions for hand-eye calibration.
- Touching uses tactile feedback for object alignment.
- Demonstrated tasks are pressing buttons and moving sliders.
- Vendors
- Unitree
- Products
- Unitree G1
From the original article
1 min read at arxiv.orgReal-to-Sim-to-Real (Real2Sim2Real) enables structured world representations for planning, interpretability, and safe execution. Most existing approaches rely on passive perception, which creates real2sim gaps due to imperfect observation. Inspired by how humans perceive actively through actions, we propose ActiveAlign, a framework that leverages active perception to align the Real2Sim2Real gap from interactions.
By Ri-Zhao Qiu · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.