mimic robotics builds dexterous humanoid robotic hands paired with standard industrial arms, driven by foundation models that learn manual tasks from human demonstrations. An ETH Zurich spin-off, it targets manufacturing, assembly, and logistics work too intricate for conventional automation.
Notable milestones from mimic robotics’s journey — funding rounds, launches, and other moments worth knowing.
mimic and Black Forest Labs paired FLUX 3 with mimic's video-action-model recipe and data to create FLUX-mimic, tested in Audi production. On a real soft-body kitting task, mimic reported 95% out-of-the-box success without task-specific fine-tuning, versus 55% for an adapted pi0.5 baseline and 70% for a post-trained flow-matching policy.
mimic unveiled the M1, a tendon-driven hand with 15 active degrees of freedom across 21 joints, and the U1 exoskeleton, which captures human demonstrations with matching kinematics, contact geometry and camera perspective. Custom low-latency middleware and Video Action Models complete the vertical stack, keeping hand morphology fixed from pre-training through factory deployment instead of translating human data to a two-finger gripper.
Sourcemimic released mimic-video, a Video-Action Model that swaps a vision-language backbone for NVIDIA's Cosmos-Predict2 video model and trains a small action decoder on top. The company reported that about 500 bimanual trajectories were enough for real-robot success, with simulated benchmarks showing 10x better sample efficiency and 2x faster decoder convergence than a pi0.5-style VLA baseline.
Elaia and Speedinvest co-led an oversubscribed round — with Founderful, 1st Kind, 10X Founders, 2100 Ventures and the Sequoia Scout Fund joining — taking total funding past $20M. mimic said it was already piloting with Fortune 500 manufacturers, global automotive brands and multinational logistics providers.
Elaia, 
Speedinvest
mimic's edge is its data engine: skilled operators wear capture devices while doing real factory work, and the recorded movements train imitation-learning models that let the hands reproduce human technique and self-correct as objects shift.
mimic took the third and final stage of Switzerland's Venture Kick, adding CHF 150,000 in non-dilutive backing to develop its robotic-transformer-driven manipulation models.
SourceFounderful led the first round, joined by another.vc, Tiny.vc and angels. The demo that sold it: a tendon-driven humanoid hand picking up a croissant and making coffee — fine-motor tasks conventional grippers can't touch.
Five founders out of ETH Zurich's Soft Robotics Lab — Stefan Weirich, Stephan-Daniel Gravert, Elvis Nava, Benedek Forrai and lab director Robert Katzschmann — started mimic on a contrarian thesis: skip the full-body humanoid and put human-level dexterity into a hand that bolts onto a conventional robot arm.
