Qianjue Robotics Raises Hundreds of Millions as Tactile Sensing Heats Up

Capital is flowing toward touch. Embodied tactile company Qianjue Robotics announced two consecutive strategic funding rounds totaling several hundred million yuan, with investors including BlueRun Ventures, funds under CICC Capital and Jingya Capital.

The rounds follow a hundred-million-yuan financing announced just two months earlier, in July. Founded in May 2024, Qianjue has drawn backing from financial investors such as Hillhouse Ventures, Plug and Play China, Gobi Partners, Oriza Seed and Futeng Capital, alongside industrial players including Li Auto, AgiBot and Jide Electric, with several institutions investing across multiple rounds.

The new capital will fund iteration and scaled delivery of visuotactile sensors, collection of real-world interaction data, construction of tactile simulation and datasets, and development of tactile embodied and tactile world models.

From sensors to a world model. Qianjue was founded by Ma Daolin, a professor at Shanghai Jiao Tong University. Ma holds bachelor's and doctoral degrees in theoretical and applied mechanics from Peking University and completed a postdoc at MIT under Alberto Rodriguez, now head of robot behavior at Boston Dynamics; in 2021 he won ICRA's sole global Best Paper Award.

Before founding the company, the team had studied tactile perception for more than a decade and was among the first to systematically propose a path of 'tactile spatial intelligence.'

Rather than remaining a single-component supplier, Qianjue extends tactile capability from perception into data and models. The stack runs from visuotactile sensors at the base, through the XTac data-collection system, Xense_Sim tactile simulation and the TacVerse dataset in the middle, to the X-TouchMind V1 tactile embodied model at the top.

The hardware core is a tri-color-light visuotactile sensor. Its contact surface is an elastic material that deforms when pressed; an embedded camera captures the deformation, and algorithms decode force, motion, slip and 3D shape information from the image. The sensor has reached ten-thousand-unit-scale production.

Beyond fingertips, palms and finger pads also contact objects during grasping. On August 25, Qianjue released its Whole-Hand Tactile Solution 2.0, keeping high-precision visuotactile sensors on fingertips while switching palms and pads to multi-zone flexible electronic skin, with an on-device tactile compute hub unifying all signals. The company says overall durability is five times that of the previous generation, with sensor contact life reaching one million cycles.

For data, real collection runs in parallel with simulation. Real data comes from the XTac UMI G1 wearable collection gripper, which captures vision, touch, pose, gripper state and IMU simultaneously and converts it directly into trainable data. Simulated data is mass-generated in virtual environments by Xense_Sim. During WAIC in July, Qianjue released TacVerse 1k, described as the first visuotactile multimodal dataset, with an initial 1,000 hours; the company says its in-house acquisition hardware cuts data cost by more than 80 percent and raises efficiency three- to five-fold.

At the model layer, X-TouchMind V1 is presented as the first VTLA (vision-tactile-language-action) embodied tactile model, unifying vision, language, touch, action and robot state. Its three tiers map to three stages of operation: System 2 handles task understanding, System 1 generates motion plans, and System 0 performs high-frequency post-contact correction, fixing slip and abnormal force in real time.

Each layer of the stack feeds the one above it, but the full-stack layout ultimately depends on whether customers buy in. To date Qianjue says it has served more than 300 paying customers across robotics, industrial manufacturing, consumer brands and research institutions.

Robotics companies including AgiBot, Google DeepMind, X Square Robot, Galbot and EngineAI purchase sensors and data-collection equipment for their own R&D, while industrial clients such as CATL, L'Oréal and Dewu want deployment solutions for specific scenarios. The former validates the technology; the latter validates the business.

Qianjue's back-to-back rounds are not an isolated case; the industry has begun calling 2026 the 'first year of touch.' PaXini closed a B round exceeding 1 billion yuan in March, added a 1 billion yuan strategic round in August and disclosed a B++ round of several hundred million yuan in September, bringing its total past 4 billion yuan, with Samsung joining the latest round.

Tashan Technology has raised three times in about half a year, including a several-hundred-million-yuan B round in July backed by industrial players such as Joyson Electronics, Taiping Innovation and AUX; the company is also Nvidia's first global tactile simulation partner.

Over the past two years, embodied AI capital has mainly flowed to robot bodies and brains. Now that bodies can walk and brains can understand scenes, robots are entering warehouses and production lines to perform real work, exposing the weak link: unstable grasping, inaccurate assembly, failed insertion — all contact problems. For investors, touch offers another appeal: they do not need to pick which body maker wins. Qianjue's customer list includes competing body makers that nonetheless buy sensors from the same company. AgiBot is also a Qianjue shareholder; industrial capital investing in component suppliers typically aims to lock in capacity early.

At the end of August, Ma published a signed article for the World Economic Forum titled 'Why tactile intelligence may become the next infrastructure layer of physical AI.' 'Infrastructure layer' captures exactly this relationship: the contest among robot bodies is far from settled, but sensors, data and models are needed by everyone.

With the body race undecided, tactile orders are already arriving.