TARS Robotics
Last reviewed
May 11, 2026
Sources
14 citations
Review status
Source-backed
Revision
v5 · 2,274 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
May 11, 2026
Sources
14 citations
Review status
Source-backed
Revision
v5 · 2,274 words
Add missing citations, update stale details, or suggest a clearer explanation.
| TARS Robotics | |
|---|---|
| General information | |
| Full name | TARS (Trusted AI and Robotics Solution); 它石智航 (Tashi Zhihang) in Chinese |
| Founded | February 5, 2025 |
| Founders | Chen Yilun (CEO), Li Zhenyu (Chairman), Ding Wenchao (Chief Scientist) |
| Headquarters | Beijing and Shanghai, China |
| Industry | Robotics, Embodied AI |
| Products | TARS A Series (industrial), T Series (general-purpose), AWE foundation model, WIYH dataset |
| Total funding | ~$697 million (approximately CNY 5 billion) by April 2026 |
| Notable record | Guinness World Record for sub-millimeter wire harness assembly (March 2026) |
| Website | tars.ai |
TARS (Trusted AI and Robotics Solution), also known by its Chinese name 它石智航, is a Chinese robotics and embodied AI startup founded in February 2025. The company develops humanoid robots and a general-purpose embodied foundation model called the AI World Engine (AWE). Despite being one of the youngest companies in the humanoid robotics sector, TARS has set multiple Chinese venture capital records, including the country's largest seed embodied AI round in March 2025 and the largest single Pre-A round in China's embodied intelligence history in April 2026, lifting cumulative funding to roughly CNY 5 billion (about $697 million) within fourteen months of incorporation.[1][2][3]
The startup is led by Chen Yilun, the former CTO of Huawei's autonomous driving unit, and Li Zhenyu, the former head of Baidu's Intelligent Driving Group and architect of the Apollo platform. TARS publicly demonstrated its first robots in December 2025, when a humanoid prototype threaded a needle and stitched a logo by hand, a feat the company described as a world first in fine-motor humanoid manipulation.[4][5]
TARS was incorporated on February 5, 2025, by three technology leaders drawn from China's largest autonomous-driving and AI programs. The company's full English name, Trusted AI and Robotics Solution, reflects its stated mission of moving artificial intelligence from screens into machines that can act in the physical world.[2][6]
| Founder | Role | Prior experience |
|---|---|---|
| Chen Yilun | CEO and co-founder | Former CTO and chief scientist of autonomous driving at Huawei; former chief machine vision engineer at DJI; chief expert on AI robotics at Tsinghua University's Institute for AI Industry Research (AIR) |
| Li Zhenyu | Chairman and co-founder | Former senior vice president at Baidu and head of its Intelligent Driving Group (IDG), where he created the Apollo self-driving platform |
| Ding Wenchao | Chief Scientist and co-founder | Former Huawei "Genius Youth" recruit (2020), led Huawei ADS prediction and decision-making team; researcher at Fudan University |
The team established dual headquarters in Beijing and Shanghai. Beijing houses corporate operations and research collaborations with Tsinghua AIR, while Shanghai serves as the engineering hub for hardware development and large-scale data collection. Internal documents and press releases use the legal entity name TARS AI for the Shanghai unit and TARS Robotics for the international identity.[1][3]
Meituan's strategic investment team described TARS as a candidate to become "the next Unitree Robotics" when explaining its decision to anchor the company's Angel+ round, reflecting how investors saw the founding trio's combined operating experience as unusually deep for a seed-stage robotics startup.[6]
TARS has raised four publicly disclosed rounds in just over a year, each described in Chinese financial media as a sector record at the time of closing.[1][3][7]
| Round | Date | Amount | Lead investors | Notes |
|---|---|---|---|---|
| Angel | March 2025 | ~$120 million (CNY 800 million) | Lanchi Ventures, Qiming Venture Partners | Reported by Chinese state media as the largest seed embodied AI investment to that point |
| Angel+ | July 2025 | ~$122 million | Meituan Strategic Investment | Meituan's first lead role in an embodied intelligence round; participants included Junshan Capital, Bihong, Guoqi, Lingang Sci-Tech, SAIF Partners, C&D, Linear Capital, Xianghe |
| Pre-A (cornerstone tranche) | Late 2025 / Early 2026 | undisclosed | Beijing Robotics Industry Development Investment Fund (state-backed) | Reported as part of the broader Pre-A wave |
| Pre-A | April 16, 2026 | ~$455 million (over CNY 3 billion) | Hillhouse (GL Ventures), HongShan (Sequoia China), Meituan Strategic Investment | Largest single-round and largest Pre-A in China's embodied intelligence history |
The April 2026 Pre-A round drew an unusually broad investor group, including CICC Capital, Capitallink, Oriental Fortune, PEAKVEST, Qiming Venture Partners, Linear Capital, Lanchi Ventures, Xiang He Capital, Hongtai Fund, TCL Capital, Futeng Capital, Shoucheng Holdings, C&D Emerging Investment, Hengxu Capital, CICV Investment, the Beijing Robotics Industry Development Investment Fund, and the Shanghai State Investment Guide Fund. Caixin and EqualOcean both reported cumulative TARS funding at approximately CNY 5 billion (around $697 million) after the round closed.[1][3][7]
The Pre-A close came just weeks after TARS's A1 industrial robot set a Guinness World Record for sub-millimeter wire harness assembly in March 2026, a sequencing that several Chinese outlets noted as evidence of how operational milestones, rather than pure technical promises, were pulling top-tier Chinese venture capital into the humanoid sector.[1][2]
TARS publicly unveiled its first two robot prototypes in December 2025, less than ten months after the company was founded. The launch event, held in Shanghai, showed an A Series industrial humanoid working alongside a T Series general-purpose humanoid that walked on two legs. Both robots shared the same upper body and AI brain but used different lower-body designs to address distinct deployment scenarios.[4][8]
At the same December 2025 event, the company opened public access to its WIYH (World In Your Hands) dataset and previewed the embroidery demonstration that drew widespread coverage in Chinese tech and international engineering press.[5][9]
TARS organizes its product portfolio around three pillars: hardware platforms (A Series and T Series), the AWE foundation model, and supporting data infrastructure (SenseHub and the WIYH dataset).
| Product | Type | Locomotion | Specifications | Target market |
|---|---|---|---|---|
| A Series (including A1) | Industrial humanoid | Wheeled or stationary base | Approximately 160 to 175 cm tall, around 80 kg, five-fingered dexterous hands, electric servo actuators with harmonic and planetary gears | Flexible manufacturing, automotive wire harness assembly, logistics |
| T Series | General-purpose humanoid | Bipedal | Walking speed around 3 km/h, three-hour runtime, 30 to 40 degrees of freedom overall, 10 to 20 in the hands | Mixed-use settings, household pilots, research |
Both platforms use the same upper body to maximize software reuse and reduce data-collection cost. The hands include real-time tactile and force feedback at sub-millimeter precision, which TARS describes as the foundation for tasks involving deformable materials such as fabric, wiring, or food. The reference price quoted for early units is around $95,000, although the company has not opened broad commercial sales as of April 2026.[10]
AWE is TARS's proprietary embodied foundation model. The company has released two public iterations:[4][9]
TARS describes AWE as the layer that turns large-scale operational data into transferable physical skills. Chief Scientist Ding Wenchao has said the model is trained for generalization, not single-task scripting, so that the same robot can be reassigned across factory lines or domestic settings without bespoke programming.[4][8]
SenseHub is TARS's data-collection platform. Rather than gathering training data in controlled lab settings, the system captures synchronized streams of RGB video, fingertip pressure signals, and finger joint movements from real workflows in hotel laundry rooms, supermarket assembly counters, and logistics warehouses. SenseHub feeds the WIYH (World In Your Hands) dataset, which TARS describes as the first large-scale Vision-Language-Tactile-Action multimodal dataset for embodied intelligence.[9]
WIYH was opened for public access by research institutions and industry partners in December 2025. Its design emphasizes four attributes: drawn from real operational tasks, spread across multiple industries, fully annotated across visual, language, tactile, and action modalities, and scaled to match the data volumes used to train large language models.[9]
CEO Chen Yilun describes the company's technical approach as a "DATA-AI-PHYSICS trinity," a closed loop connecting real-world data collection (SenseHub), generalist embodied models (AWE), and physical execution on the A and T Series robots. The stated goal is to minimize what TARS calls the digital-to-physical gap, the loss in capability that typically occurs when policies trained in simulation are deployed on real hardware.[4][5]
The trinity model has shaped TARS's choice to invest heavily in its own data pipeline rather than relying on synthetic data or publicly available robot demonstration datasets. The company argues that authentic data from messy workplaces is what allows the same model to drive both an industrial wire-harness routine and a domestic-style bimanual sewing task.[4][8]
TARS's hands include continuous real-time tactile and force feedback at the fingertip. In the December 2025 demonstration, the robot threaded a sewing needle and stitched the TARS logo onto fabric using both hands in coordination. Press releases and third-party reporting framed this as a world first for humanoid robots, because needle-threading combines sub-millimeter precision, deformable-material handling, and bimanual coordination, three properties that flexible manufacturing has historically been unable to automate.[4][5][8]
In March 2026, the A1 industrial robot extended this dexterity to a high-volume task by setting a Guinness World Record for "the most sub-millimeter wire harnesses assembled by a robot in one hour," completing 105 sub-millimeter precision wiring harness assemblies inside the hour. The record was noted as the first Guinness honor in industrial precision operations for a Chinese embodied AI company.[3][7]
TARS targets manufacturing tasks that have resisted conventional automation, especially those involving flexible or deformable materials. Wire harness assembly, which involves routing soft wires through dense connector ports without tangling, has been called the last major bottleneck in automotive and aerospace manufacturing, and TARS positions the A Series as the first robotics product to address it commercially.[2][7]
The T Series is aimed at longer-horizon scenarios, including general-purpose tasks in mixed indoor environments. Investor briefings reported by 36kr emphasize that Meituan, with its delivery and retail operations, views embodied robots as a strategic complement to its existing logistics network, and that the company has been exploring TARS robots for in-store and back-of-house tasks.[6]
As of April 2026, TARS has not announced mass commercial availability of either robot line. Product references suggest pilot deployments and an indicative early-unit price around $95,000, with full commercial launch timing tied to AWE 3.0's industrial rollout and the maturation of the SenseHub data pipeline.[10]
Chinese venture capital and policy media have positioned TARS as a marker of how quickly capital and talent have migrated from autonomous driving into humanoid robots. The founders' overlapping experience at Huawei, Baidu, and DJI has been cited as the main reason TARS attracted top-tier investors in its seed round without a working prototype.[1][6]
International coverage has focused on two episodes in particular. The December 2025 embroidery demonstration was reported by Interesting Engineering, Robotics Business News, The AI Journal, and other outlets as evidence that fine-motor humanoid skills had reached a level previously assumed to be years away. The March 2026 Guinness record framed TARS as a manufacturing-grade automation supplier rather than only a demo-focused research lab, and helped justify the record-breaking $455 million Pre-A round that closed weeks later.[3][5][8]
TARS's rise is also frequently contrasted with that of Unitree Robotics and Figure AI. Unlike Unitree, which began with quadrupeds and added humanoids later, TARS launched directly into bimanual humanoid hardware tied to an embodied foundation model. Unlike Figure AI, which has emphasized partnerships with U.S. manufacturers, TARS has anchored its commercial story in Chinese flexible manufacturing supply chains and Chinese state-backed industrial funds.[6][7]