TARS它石智航
LatestPre-A (Apr 16, 2026, $455M)
Undisclosed (top tier)
Shanghai TARS Technology Co., Ltd.
Non-VLA world-model route (AWE AI World Engine); 'no VLA, no simulation, no teleoperation'; human-centric data paradigm; AWE 3.5 won the WAIC 2026 SAIL Star; China's first 1,000-unit industrial embodied robot cluster landed in Shanghai Jiading; $455M set China's embodied single-round record
数据截至2026-08-19
About
TARS was co-founded by Chen Yilun (CEO), former Huawei autonomous driving CTO and chief scientist of its Intelligent Automotive Solution BU; Li Zhenyu (Chairman), former head of Baidu's Intelligent Driving Group; and Ding Wenchao (Chief Scientist), one of Huawei's first 'Genius Youth' cohort (registered July 2024, formally operating from February 2025) — called one of the strongest founding teams in embodied intelligence. The company insists on a non-VLA route, self-developing the AWE (AI World Engine) world model, proposing 'no VLA, no simulation, no teleoperation' and a human-centric data paradigm, with over 1 million hours of cumulative data and a push toward 10 million hours by end-2026. AWE 3.0 is the world's first general embodied foundation model that 'can actually work' and set the Guinness World Record for most sub-millimeter wire-harness assemblies in one hour; AWE 3.5 won the WAIC 2026 SAIL Star. Commercialization focuses on industrial scenarios such as automotive wire-harness assembly: with THB Electronics it drives the world's first scaled deployment in automotive wire harnesses; it deployed 100 robots at an Aptiv plant to build China's first 100-unit 'real work' cluster; and China's first 1,000-unit industrial embodied robot cluster landed in Shanghai Jiading. Its $242M angel and $455M Pre-A rounds broke industry funding records back to back, totaling over RMB 4.6B within 14 months of founding.
Technical Approach
Explicitly rejects the VLA route, self-developing the AWE (AI World Engine) architecture to realize embodied intelligence as a world model, insisting on 'no VLA, no simulation, no teleoperation', driven by real physical-world data. It champions a human-centric data collection paradigm, turning human operation control, force perception and real-time decision experience into learnable data assets (over 1 million hours cumulative, pushing for 10 million by end-2026). AWE 3.5 is the first embodied-native model to realize the complete 'pre-training + post-training' paradigm, with a methodology loop of 'native architecture + dual-prior pre-training + world-model-driven post-training + continuous data feedback'.
Commercialization（6）
- 1,000-unit industrial cluster: China's first 1,000-unit industrial embodied intelligence robot cluster landed in Shanghai Jiading, to be deployed through 2027 (per the Shanghai Municipal Commission of Economy and IT, 2026-07-10)
- Aptiv partnership: deployed 100 robots at an Aptiv plant in 2026, building China's first 100-unit industrial 'real work' robot cluster; covering 8 major production processes, 33 application scenarios and 46 core production stations (per People's Daily Overseas Edition)
- In June 2026 signed a strategic cooperation agreement with domestic wire-harness leader THB Electronics, driving the world's first scaled landing of embodied intelligence in automotive wire harnesses
- The embodied intelligence demonstration project officially landed in Shanghai, using Shanghai as the core industrialization base to replicate across the Yangtze River Delta and nationwide (per the Shanghai Commission of Economy and IT)
- WAIC 2026: a 1:1 replica of a real automotive wire-harness factory line, with the robot cluster performing autonomous production-line work live; the industry's first public real-time interaction between humans and embodied models
- Strategy of 'Foundational Model + App', taking vertical applications into real industrial sites; judges that 'manufacturing will become embodied intelligence's most important direction over the next three years'
Position & Strengths（5）
- A top founding team: Huawei autonomous driving CTO (Chen Yilun) + Baidu IDG president (Li Zhenyu) + one of Huawei's first 'Genius Youth' (Ding Wenchao) — arguably one of China's strongest embodied founding teams
- Non-VLA differentiated route: a self-developed AWE world model, no VLA, no simulation, no teleoperation — a unique technical path; AWE 3.5 won the WAIC SAIL Star
- Funding record holder: the $242M angel and $455M Pre-A set China embodied records back to back, over RMB 4.6B in total
- First in industrial landing: China's first 1,000-unit industrial cluster, the Aptiv 100-unit 'real work' cluster, and the world's first scaled automotive wire-harness deployment with THB
- Data assets: over 1 million hours of human-centric data, pushing for 10 million hours by end-2026