🔥36氪•Stalecollected in 8m
Industrial Physical AI: From Perception to World Action
#industrial-ai#embodied-ai#world-modelsjx-phi-world-/-jx-phi-brainjiangxing intelligencedeepseekqwen
💡Learn how industrial AI is moving beyond simple vision tasks to complex, multi-step physical world execution.
⚡ 30-Second TL;DR
What Changed
Introduced JX-Phi World for data processing and simulation (AutoEdge/AutoWorld).
Why It Matters
This architecture provides a blueprint for deploying AI in high-stakes industrial environments where reliability and safety are paramount.
What To Do Next
Evaluate your industrial use case for 'long-task' decomposition potential using VLA models to improve automation reliability.
Who should care:Enterprise & Security Teams
Key Points
- •Introduced JX-Phi World for data processing and simulation (AutoEdge/AutoWorld).
- •JX-Phi Brain evolves towards World Action Models (WAM) using S-VLM and LT-VLA.
- •Focuses on high-reliability industrial tasks requiring multi-step planning and execution.
- •Leverages China's unique industrial density and 5G infrastructure for scale.
🧠 Deep Insight
Web-grounded analysis with 12 cited sources.
🔑 Enhanced Key Takeaways
- •Jiangxing Intelligence, established in 2018 and headquartered in Nanjing, China, has expanded its operational footprint with research and development centers and marketing teams in Shenzhen and Beijing.
- •The company has successfully completed five rounds of financing, with its most recent being a Series B++ funding round in March 2026, led by SparkEdge Capital, with participation from CD Capital, JinkoSolar Holding Co., Ltd., Yankuang Capital, Xiamen C&D Emerging Industry Equity Investment, and G&O.
- •Jiangxing Intelligence's solutions have demonstrated tangible economic benefits, such as reducing maintenance expenses by 60% in a Sichuan hydropower plant, resulting in annual savings of tens of millions of yuan.
- •The company's core offering includes the 'IDEA edge computing platform,' an industrial-grade intelligence system that facilitates AI technologies and digital transformation for clients, providing customized services like transmission line inspection and virtual power plant scheduling.
- •Jiangxing Intelligence is an active partner of EdgeX Foundry, utilizing EdgeX-based solutions to enhance sensor connectivity, data access, processing, integration, and analysis for its AI IoT devices and vertical solutions in sectors such as water affairs and smart grids.
🛠️ Technical Deep Dive
- Three-Layer Full-Stack Architecture: Jiangxing Intelligence proposes an architecture that integrates data infrastructure, world models, and agent-based execution for complex industrial tasks.
- JX-Phi World: This component is responsible for data processing and simulation, leveraging capabilities like AutoEdge and AutoWorld.
- JX-Phi Brain: This evolves towards World Action Models (WAM), incorporating Spatial Vision-Language Models (S-VLM) and Long-Term Vision-Language-Action (LT-VLA) capabilities.
- World Models (WMs): These are AI systems designed to predict how an environment will evolve based on current states and agent actions, effectively providing an internal 'physics engine' and 'process engine' learned from data. They integrate multimodal data, including text, images, audio, sensor signals, and video sequences, across spatial, temporal, and physical dimensions to represent the real world.
- Physical AI Principles: The system integrates physical modeling, AI algorithms, and industrial domain knowledge, utilizing massive multimodal data (e.g., visual images/videos, acoustic signals, vibration sensor data, temperature/humidity time series, production process text records) for accurate perception, reliable decision-making, and intelligent control in industrial processes.
- Vision-Language-Action (VLA) Models: These are critical for autonomous decision-making in embodied systems, often employing textual planning (Chain-of-Thought reasoning) before executing low-level actions. Advanced VLA systems, such as MindExplore, feature a hierarchical embodied intelligence architecture with reasoning, acting, and memory components, capable of adapting to dynamic environments using multimodal sensor inputs like RGB, depth, and LiDAR data.
- Runtime Reasoning-Action Alignment Verification: To enhance robustness and enable novel behavior composition, Jiangxing's approach likely incorporates mechanisms for verifying the alignment between textual plans and generated actions, potentially using a pre-trained Vision-Language Model (VLM) as a critic during runtime.
- Edge Computing Platform: The company offers cloud-edge-end collaborative industrial-grade intelligent systems, including the IDEA edge computing platform, which supports customized services.
- EdgeX Foundry Integration: Jiangxing Intelligence leverages EdgeX Foundry for efficient sensor connection, data access, processing, integration, and analysis in its AI IoT solutions.
🔮 Future ImplicationsAI analysis grounded in cited sources
Jiangxing Intelligence will significantly expand its market share in high-value industrial sectors like mining, chemical, and rail transit beyond its current energy and power focus.
The company explicitly stated its intention to use recent funding to accelerate the replication and expansion of successful solutions from energy and power to these high-value scenarios.
The adoption of Jiangxing Intelligence's physical AI solutions will lead to a measurable increase in operational efficiency and a reduction in unplanned downtime across various Chinese industrial sectors.
Physical AI systems, particularly those using world models, are designed to overcome the brittleness of traditional automation by enabling machines to understand and simulate physical world behavior, leading to more robust and adaptable operations, and Jiangxing has already demonstrated significant cost savings in early deployments.
Jiangxing Intelligence's emphasis on 'World Action Models' (WAM) and advanced Vision-Language-Action (VLA) will position it as a key enabler for the next generation of autonomous industrial systems in China.
World models and VLA are considered foundational for physical AI systems that can perceive, reason, and act in complex, dynamic real-world industrial environments, moving beyond static perception and rule-based control.
⏳ Timeline
2018
Jiangxing Intelligence (Nanjing Jiangxing Lianjia Intelligence Technology Co., Ltd.) was established in Nanjing, China.
2020-11
Completed Series A1 funding round.
2021-06
Secured Series A++ financing round led by Zhongguancun Innovation Fund, Future Capital, and Zhuoyuan Capital, with existing shareholders including Sequoia Capital, Baidu Ventures, and Lenovo Ventures.
2022
Selected for the 2022 Sci-Tech China Emerging Enterprises List and recognized as User Satisfaction Supplier of the Year.
2022-08
Completed a Series Pre-B funding round worth hundreds of millions of CNY, led by Everest VC.
2026-03
Raised a Series B++ funding round led by SparkEdge Capital.
📎 Sources (12)
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: 36氪 ↗