Launched at ICRA 2026 by Daimon Robotics and Galbot, RobOmni gives the field its first shared way to grade how touch changes manipulation, and a ruler the community now has to challenge.
For years, the robotics community has run on a quiet assumption: that adding tactile sensing to a robot's perception stack improves physical interaction. The assumption makes intuitive sense. Skin, slip detection, and force feedback help humans manipulate objects they cannot see. Robots should benefit similarly.
The trouble is that nobody had a clean way to measure whether the assumption is true, or how much, or under which conditions. Vision-led benchmarks have proliferated across embodied AI. Tactile has not. So claims about "tactile intelligence" have floated untethered from comparable evidence.
That is the gap RobOmni, reported by The Robot Report, is now trying to fill. Launched at ICRA 2026 by Chinese-headquartered robotics developers Daimon Robotics and Galbot, RobOmni is described as the first omni-modal evaluation benchmark to include tactile perception alongside vision and language.
The framing matters. A benchmark is not a demo. A demo shows a robot doing something impressive once. A benchmark gives the community a shared ruler: the same tasks, the same scoring, the same contact conditions, run across systems so results can actually be compared. Without that ruler, "tactile helps" and "tactile does not help" become unfalsifiable claims. With it, progress becomes legible.
According to The Robot Report's coverage, RobOmni targets a specific set of manipulation problems: grasping, insertion, assembly, object placement, and tool handling. It also targets the contact conditions that distinguish a tactile benchmark from a vision-only one: slip, force, deformation, material, geometry, and texture. Those are the variables where touch should in principle add information that cameras cannot recover.
Daimon has cast the broader framing as "omni-modal tactile intelligence," a category the company is using to position its platform. RobOmni is the testable surface of that positioning. If omni-modal tactile intelligence is a real category, the benchmark is where researchers, integrators, and buyers can start probing it. If the category does not hold up, the benchmark should expose that on its first round of independent submissions.
A few honest caveats belong in any first read of the launch. The Robot Report is an industry-vertical trade outlet republishing a company-led announcement, so the framing of the gap and RobOmni's novelty is the launching companies' view rather than settled consensus. No primary paper, arXiv preprint, or project page has been independently verified here, which means the technical details of the benchmark design, baseline tables, and submission policy remain to be checked against the original release. The headline phrase "jointly launches" may also overstate the depth of the collaboration. A shared benchmark with separate contributions is a different kind of partnership than a co-developed artifact, and the primary sources will tell readers which one this is.
None of that undercuts the underlying news. A first omni-modal benchmark that includes tactile is, on its face, the kind of infrastructure the field has been missing. The question is whether the community treats it as a starting point or as a verdict. If the design surfaces what integrators and researchers actually need to compare, and if it stays open to revision after the first round of submissions, it has a chance of becoming the shared yardstick the launch implies. If it does not, the next attempt at this kind of benchmark will be built by whoever is unhappy with the first one.
What to watch next: the primary paper or arXiv release, the code and dataset policy, the first independent submissions from groups outside Daimon and Galbot, and whether competing tactile-manipulation benchmark efforts choose to align with RobOmni's task design or to fork.