千诀科技
Category: Robotics / Embodied AI
Tsinghua-spinout building distributed predictive world models (embodied brain) for autonomous robot decision-making and planning 千诀科技 was founded in 2023. The company is led by 高海川 (Gao Haichuan). Based in Beijing, China. Team size: 101-500. Total funding raised: $100.0M. Latest round: Series A. Key investors include 京铭资本, 英诺天使基金, 祥峰投资 (Vertex Ventures), 钧山投资, 德同资本, 追创创投 (Dreame Innovation Capital), 华业天成, 瑞江投资, 石溪资本 (GigaDevice), 水木清华校友种子基金, 启迪之星创投, 山东新动能.
- Founded
- 2023
- Headquarters
- Beijing, China
- Team size
- 101-500
- Total funding
- $100.0M
Value proposition
Provides a universal "embodied brain" that enables any robot morphology to autonomously perceive, plan, and act in dynamic real-world environments without human teleoperation, pre-programming, or environmental modifications
Products and solutions
Qianjue Embodied Brain (具身大脑) - robot perception & decision-making large model, Distributed Predictive World Model (分布式预测世界模型), Polibrain OS - robot brain operating system, Brain Dock (脑坞) - edge computing hardware based on Tianjic chips supporting 13B parameter models
Unique value
Brain-region-inspired distributed predictive world model architecture that decouples perception/planning from execution, enabling cross-robot-platform transferability with 10x sample efficiency vs. generative approaches, running on fully domestic edge hardware
Target customer
Robot manufacturers (humanoid, wheeled, quadruped, drone, cleaning robot makers); Commercial service operators (hotel cleaning, restaurant service, precision indoor operations)
Industries served
Robotics, Commercial Services, Hospitality (hotel cleaning), Food Service (restaurant delivery), Smart Cleaning, Precision Indoor Operations, Smart Manufacturing
Technology advantage
Predictive world model (vs. generative); Brain-region-partitioned distributed architecture; Brain-cerebellum decoupling for cross-platform transfer; Fully domestic edge hardware (Tianjic chip-based Brain Dock) supporting 13B models; 100,000+ deployed terminal devices generating real-world data flywheel; Spatio-Temporal Approximation (STA) method published at ICLR 2024 enabling Transformer-to-SNN conversion without retraining
How they differentiate
Unlike mainstream generative/pixel-reconstruction world models, Qianjue uses a predictive world model that learns low-dimensional physical state evolution trajectories (not pixel reconstruction), avoiding "feature contamination." Their brain-region-partitioned architecture (类脑分区) mimics human brain functional areas for distributed compression and prediction, achieving higher sample efficiency and faster inference. The brain-cerebellum decoupling design allows the same brain to transfer across robot morphologies without retraining.
Main competitors
Physical Intelligence (US - primary benchmark), 星海图 (Xinghaitu), 自变量机器人 (Zibianliang Robot), 智平方 (Zhipingfang), 千寻智能 (Qianxun Intelligence)
Key partnerships
Tsinghua University Brain-like Computing Research Center (清华大学类脑研究中心), Tsinghua VIPLAB (自动化系VIPLAB), Dreame Technology ecosystem (追觅科技), Maple Pledge枫承资本 (long-term PE/VC financing advisor)
Notable customers
Multiple internet/3C giants (unnamed per company), Dreame Technology (追觅科技) ecosystem companies, Partners in hotel cleaning, restaurant service, and commercial cleaning sectors
Major milestones
2023-06: Company founded, incubated from Tsinghua Brain-like Computing Center, 2024-04: Released first product-grade robot perception & decision-making large model, 2024-11: Completed Angel round (数千万元), 2025-01: Completed Angel+ and Angel++ rounds (累计亿元级), 2025-03: Completed Pre-A round, 2025-05: Completed Pre-A+ round (数亿元), 2025-12: Completed Pre-A++ round (近亿元), 2026-02: Completed Pre-A++ extension round, 2026-06: Completed Series A (数亿元, led by 京铭资本), 2026-06: Achieved 100,000+ deployed terminal devices across multiple robot types
Growth metrics
100,000+ terminal devices deployed; 3 generations of embodied brain pre-trained; Billion-scale embodied perception & decision pre-training dataset; Cross-robot compatibility across wheeled, quadruped, bipedal humanoid, drone, and cleaning robot platforms
Market positioning
Positioned as the leading "embodied brain" provider in China, directly benchmarking against US-based Physical Intelligence. Differentiated by brain-region-partitioned architecture and predictive (vs. generative) world model approach. Targeting the mass-market robot brain OS layer, analogous to Android for smartphones.
Geographic focus
China (primary), with plans for global commercialization
Patents and IP
Spatio-Temporal Approximation (STA) method published at ICLR 2024 - first training-free method for converting Transformers to spiking neural networks; Multiple publications at NeurIPS, ICML, AAAI, ICLR, ICCV, TPAMI on world models, representation learning, and robot learning
About 高海川 (Gao Haichuan)
PhD from Tsinghua University, Department of Automation; Since 2018, led the brain-like dual-arm robotics research group at Tsinghua Brain-like Computing Center as group leader, designing multiple dual-arm autonomous decision-making robots from scratch
Latest news about 千诀科技
- Qianjue Technology (千诀科技) raises hundreds of millions of yuan in Series A for embodied intelligence world models
- Qianjue Technology secured nearly RMB 100 million in Pre-A++ funding to scale its full-size household robots. By utilizing pure-vision spatial understanding instead of LiDAR, the startup aims to conne
More Robotics / Embodied AI companies
Official website: https://www.qj-robots.com
