Robot funding is moving from whole machines into the fingers. On August 11, Daimon Robotics said it had completed a strategic round worth hundreds of millions of yuan, led by Ant Group with existing investors adding capital.
QbitAI framed the deal as Ant's first move into tactile perception after earlier bets across humanoid robots, dexterous hands and embodied AI companies. Beijing News confirmed the financing and added the harder commercial metrics: Daimon began formal operations in Shenzhen in 2023, has shipped vision-based tactile sensors at the ten-thousand-unit level, serves more than 200 customers, has over 50 overseas customers, and has delivered to organizations including OpenAI, Figure, BMW, Google DeepMind and NUS.
The missing layer is contact
The story is larger than one financing round. Daimon recently introduced Daimon-TWM, a tactile-grounded world model that links tactile sensing, real interaction data and robot control. QbitAI quoted Carnegie Mellon professor Matthew T. Mason: “The dexterity of a hand is mostly about the brain, not the hand.”
In practical terms, vision can see a shard of glass, but touch tells the robot whether the shard is slipping, whether the tablecloth is being pulled up, and whether more force will crush the object. Daimon's 3D strategy, shown on its own site, is Devices, DaaS and Deployment: sell tactile hardware, collect real-world interaction data, then feed the data back into models and robot deployments.
A 10B-class model plus tactile data
QbitAI says Daimon-TWM uses physical cognition, planning and instant control modules. The model is described as 10B-class and able to run in real time on an NVIDIA RTX 5090 platform, with the fast-control layer correcting actions at 100Hz. Daimon also claims average success rates in contact-heavy tasks are two times higher than pi0.5 without disturbance and ten times higher with disturbance. Those are company-side evaluation claims, not a third-party general leaderboard.
The dataset evidence is easier to verify. ModelScope lists Daimon-Infinity under a CC BY-NC-SA 4.0 license and shows more than 4.62 million downloads. Its README says the project plans to open more than 10,000 hours of vision-tactile-language-action data, while the current overview describes the first 1,000-hour release and statistics based on about 1,209 collected hours. It also lists 30 FPS video, Parquet files, a 114-dimensional observation.state, a 111-dimensional action field, and tactile deformation, shear and depth streams.
“To drive the advancement of the entire embodied AI field, we have open-sourced 10,000 hours of the dataset for the broader community.”
The bet is an infrastructure bet
IEEE Spectrum's interview adds another anchor: Daimon's fingertip module packs more than 110,000 effective sensing units. The company is selling sensors, but the strategic prize is the data entrance those sensors create.
RobOmni points in the same direction. The Robot Report's sponsored article says Daimon and Galbot launched RobOmni at ICRA 2026 as a tactile-aware physical interaction benchmark built on NVIDIA Isaac Sim, covering grasping, placement, precision insertion and component assembly, with ablation tests for policies with and without tactile input.
That benchmark still needs outside adoption before it can be treated as a neutral standard. The near-term test is concrete: whether Daimon-TWM can reproduce its success-rate gains in third-party contact tasks, whether RobOmni attracts external teams, and whether the open data trains policies that transfer across robot bodies and scenes.
Sources: QbitAI, Beijing News, Daimon Robotics website, ModelScope Daimon-Infinity, CocoLoop, IEEE Spectrum, The Robot Report; verified financing scope, Ant-led round, tactile-sector positioning, ten-thousand-unit shipments, 200-plus customers, 50-plus overseas customers, Daimon-TWM 10B-class model, 100Hz control frequency, ModelScope downloads and dataset field scope.