AIBRIEF
MoSense's tens of millions of yuan angel round from Sequoia China, Hillhouse, and humanoid maker AgiBot funds a full body tactile skin and a world model (learned physics predictor) stack aimed at the simulation to reality gap.
Robots trained in simulation can lift boxes in a digital warehouse, then drop them the moment a real floor tilts. Shanghai startup MoSense just raised a tens-of-millions-of-yuan (low single-digit millions of US dollars) angel round to attack the problem where it breaks: at the surface of the robot.
The round is backed by Sequoia China, Hillhouse Venture Capital, and humanoid-maker AgiBot. The cap table pairs a top US-dollar fund with a humanoid OEM, and signals that the bottleneck has moved upstream from models to sensors.
The flagship is MoSkin, a flexible multimodal tactile skin built on electromagnetic metamaterial mechanics. It covers hands, limbs, torso, and soles, turning a robot's rigid body into a continuous six-dimensional force field. A companion world-action-tactile prediction model uses high-frequency touch to correct low-frequency control decisions, the bridge across the Sim-to-Real gap, per the company.
In founder Q&A, CEO Yan Chaoxu laid out a three-phase roadmap: safe boundary sensing (anti-collision, anti-pinching, fine manipulation); complex whole-body tasks (carrying a large box without dropping it, using a hip to push a door); and high-precision human contact such as eldercare, where he pegs the requirement at 99.999%+ success, since a one-in-100,000 failure, meaning dropping a person, is unacceptable.
The 99.999% figure is a stated goal, not a measured result, and tactile-skin startups face a long road from demo to deployable product. Watch item: whether MoSkin ships inside a real AgiBot unit this year, and whether the world-model stack holds outside controlled labs.