Force-Grounded, Cross-View
Articulated Manipulation

Providing robots with the ability to manipulate articulated objects (doors, drawers, tools, containers) remains a central challenge in robotics. A key bottleneck is the lack of large-scale, multimodal datasets that simultaneously capture what is seen, what is done, and what is felt during real physical interaction. Especially touch, tactility, and force feedback are so far underrepresented in available datasets and methods, yet critical for robust robotic deployment.

This workshop brings together the manipulation, egocentric vision, and robot learning communities to discuss emerging challenges in force-grounded, cross-view articulated manipulation. As a focus point, we host a public challenge based on the Hoi! dataset, which directly addresses challenges in embodiment transfer and force-grounding by providing synchronized visual, force, and tactile streams across human and robot embodiments.

We welcome submissions on topics related to force-grounded manipulation, tactile sensing, and interaction understanding. The workshop accepts full-length papers (8 pages) and extended abstracts (4 pages), excluding references, in ECCV 2026 format. Authors of accepted submissions will be invited to present at the poster session. Accepted full-length papers will be included in the workshop proceedings. Submissions must be anonymized for double-blind review.

The workshop challenge is centered around the Hoi! dataset, the first large-scale benchmark to jointly capture visual, force, and tactile information across human and robot embodiments during articulated object manipulation. It contains 3,048 sequences spanning 381 articulated objects across 38 indoor environments, recorded under four embodiments. The challenge evaluates how well methods leverage cross-view and multimodal signals to understand and predict articulated manipulation.