PsiBot AI灵初智能
LatestAngel + Pre-A totaling RMB 2B (2026)
~RMB 8B (industry estimate for Q1 2026; valuation multiplied six- to seven-fold within a year and a half of founding)
Beijing PsiBot Intelligent Technology Co., Ltd.
Embodied manipulation + world action model route; trains embodied models via 'body-free data collection'; released Psi-R2/Psi-W0 and open-sourced a 1,000-hour full-modality dataset of human hand manipulation
数据截至2026-08-19
About
Founded in September 2024 and headquartered in Beijing, PsiBot AI is a general embodied AI company focused on dexterous manipulation and world action models, building a large-scale data collection system to train embodied models through 'body-free data collection.' The company combines human ego data (nearly 100,000 hours) with real-robot data (5,417 hours) and covers complex scenarios through the PSI framework. In 2026 it completed Angel and Pre-A rounds totaling RMB 2B, with funds directed to commercialization scenarios such as logistics and retail and to a large-scale data collection system; it then released the embodied manipulation model Psi-R2 and the world action model Psi-W0, and open-sourced the first batch of a 1,000-hour full-modality dataset of human hand manipulation, easing the industry's 'data drought.' Mass production of the psi-SynRobot dexterous manipulation robot has started, focused on logistics scenarios. Alongside 星海图, X Square Robot and 智平方, the company belongs to the 'large models and data' competitive camp, differentiated by its focus on dexterous manipulation — regarded as the 'hardest problem' of embodied AI.
Technical Approach
Acquires human manipulation data via 'body-free data collection', without depending on its own robot bodies; runs a dual engine of human ego data (nearly 100,000 hours) + real-robot data (5,417 hours), training for complex scenarios through the PSI framework; uses the 62+ DoF tactile glove psi-SynHand for high-quality dexterous manipulation data collection, building an integrated 'simulation - real robot - internet' data collection paradigm.
Commercialization（3）
- Mass production of the psi-SynRobot has started, focused on dexterous manipulation needs in logistics (source: Zhipu panorama report)
- Commercialization scenarios focus on logistics, retail and other directions with clear task boundaries
- Data collection system construction: a large-scale human data collection network
Position & Strengths（4）
- A representative of the data route: belongs to the 'large models and data' competitive camp alongside 星海图, X Square Robot and 智平方
- Open-source contribution: open-sourced a 1,000-hour full-modality dataset of human hand manipulation, easing the industry's 'data drought'
- Differentiation: focuses on dexterous manipulation — regarded as the 'hardest thing' in embodied AI
- Ego data pioneer: pre-trains mainly on human first-person data (PSI framework)