Striding AI, a Beijing-based robotics company, announced on June 28 that it is developing a new generation of robotic foundation systems designed to put AI into physical environments. For IT and development teams watching the convergence of AI and robotics, the move signals a push toward infrastructure that connects foundation models directly to real-world action and feedback loops.
The company's approach combines advanced foundation models with robotic perception, control systems, and real-world action data. The goal is to enable robots to learn from interaction and improve over time. World Action Models and next-generation reinforcement learning form the technical backbone of this effort.
"We believe that breakthroughs in Physical AI emerge from the continuous co-evolution of data, models, and infrastructure," said Song Yao, founder and CEO of Striding AI.
A systems-first approach to physical AI
Striding AI treats physical AI as a full-stack challenge. Its architecture brings together foundation models, robot hardware and software, data infrastructure, control systems, and deployment engineering. The leadership team includes veterans from AI chip design, autonomous driving, robotics research, and industrial technology - combining research depth with production experience.
The company plans to start with structured retail environments. Robots will handle shelf restocking, inventory counting, product organization, and checkout assistance. These settings offer frequent human interaction, repeatable workflows, and rich operational data, making them a practical testbed for scaling physical AI.
Closing the loop with reinforcement learning
Behind the deployment strategy is a closed-loop robotics architecture that spans perception, planning, execution, feedback, and recovery. A key component is human-in-the-loop reinforcement learning, which turns real-world operations into continuous training data. In early internal tests, this method improved task success rates by up to 3x.
To scale that flywheel, Striding AI is building infrastructure for robot pretraining, distributed reinforcement learning, and edge-to-cloud orchestration. The platform is designed to improve as more robots operate in real environments. This infrastructure work draws on techniques familiar to teams working with Generative AI and LLM systems, but extends them into the physical domain.
From retail to broader applications
Striding AI expects its robotic foundation systems to eventually support sectors beyond retail, including food, agriculture, logistics, healthcare, and telecommunications. The long-term vision is to build robots that learn from real-world experience, improve continuously, and integrate into everyday human environments. The capabilities developed in retail - handling diverse objects, understanding store shelves, planning complex tasks - are part of an integrated system designed for broader robotic applications.
Why this matters for IT and development professionals
The announcement highlights a shift toward infrastructure that links AI models to physical action. IT and development teams will need to manage pipelines that span from cloud-based model training to edge devices operating in real time. The 3x improvement from human-in-the-loop RL underscores the value of continuous data feedback - a principle that mirrors MLOps best practices. As companies like Striding AI build the full stack, the tools and platforms for deploying physical AI will likely evolve, creating new demands for infrastructure engineering, data management, and model operations skills. For professionals already working on AI for IT & Development, the robotics domain represents a natural extension of the same core challenges.
Your membership also unlocks:
