PsiBot Raises $280M for Embodied AI

💡#280M fund boosts embodied AI data infra for robotics logistics scaling
⚡ 30-Second TL;DR
What Changed
Secured $280M in angel and Pre-Series A funding
Why It Matters
This major funding underscores surging interest in embodied AI for real-world applications like logistics. It positions PsiBot to provide valuable datasets, accelerating robotics development for AI practitioners.
What To Do Next
Explore PsiBot partnerships for accessing embodied AI logistics datasets.
Key Points
- •Secured $280M in angel and Pre-Series A funding
- •Backed by state-backed funds and strategic investors
- •Scale embodied AI data collection infrastructure
- •Expand deployments in logistics sector
🧠 Deep Insight
Background and context from public sources — not the original article. 6 sources cited.
🔑 Enhanced Key Takeaways
- •PsiBot was founded on September 1, 2024, by Dr. Viktor Wang, who has nearly two decades of experience in mobile devices, smart speakers, and robotics, with co-founder Yuanpei Chen, a post-2000s robotics prodigy trained at Stanford under Fei-Fei Li and Karen Liu[1][2][5].
- •The company developed custom exoskeleton devices for 1:1 real-world physical data collection, enabling supervised fine-tuning (SFT) and real-world reinforcement learning to optimize model performance beyond pretraining[3].
- •PsiBot was the only Chinese embodied AI company invited to the 2025 Global Developer Pioneer Conference, where it unveiled its VLA foundation model and presented breakthroughs in multimodal penetration and long-horizon skill chaining[3].
- •R1 model demonstrated 30-minute sustained reasoning and manipulation by playing live Mahjong with humans, using 'Chain of Action Thought' (CoAT) integrating perception, reasoning, and execution via an Action Tokenizer[5].
🛠️ Technical Deep Dive
- •Hierarchical end-to-end RL framework: Psi-P0 planning model for task decomposition, reasoning, and generalization using large models to understand action-environment interactions; Psi-C0 control model for execution[1][2].
- •Psi-R0, R0.5, R1: Industry-first end-to-end RL embodied models achieving long-horizon tasks with dual-handed dexterous manipulation, strong generalization across objects/scenarios, and full perception-to-action loops[1][3][5].
- •Data pipeline: Custom exoskeletons for real-world data collection, SFT fine-tuning, followed by real-world RL optimization; supports lifelong learning via memory structures[2][3].
- •R1 architecture: Separates planning/control connected by Action Tokenizer; enables Chain of Action Thought (CoAT) for extended tasks like 30-minute Mahjong with strategic reasoning[5].
- •Psi-P0 surpasses OpenAI's VPT and Nvidia's MineDojo in task complexity and accuracy for open-world planning[2].
🔮 Future ImplicationsAI analysis grounded in cited sources
⏳ Timeline
📎 Sources (6)
Factual claims are grounded in the sources below. Forward-looking analysis is AI-generated interpretation.
Weekly AI Recap
Read this week's curated digest of top AI events →
👉Related Updates
AI-curated news aggregator. All content rights belong to original publishers.
Original source: Pandaily ↗
This is a summary, not the original. Read the source, or get the weekly briefing.
The weekly digest
One email a week. Unsubscribe anytime.



