Beyond the Bot: Striding AI's Plan to Build the Brains for Physical AI

BEIJING – June 25, 2026 – A new company has emerged from stealth today with a war chest and a mission that looks past the robot itself to the invisible systems that will make it truly intelligent. Striding AI, a Beijing-based startup, announced its formal launch with a staggering angel funding round of nearly $100 million, signaling a monumental bet on what it calls "robotic foundation systems" designed to power the next wave of Physical AI.

The company’s ambitious goal is not merely to build another automaton, but to architect the entire digital and data infrastructure that allows intelligent machines to perceive, reason, and act effectively in the messy, unpredictable physical world. With backing from major corporate investors like CP Group and Huaqin Technology, Striding AI is positioning itself as a critical infrastructure provider for an era where AI steps out of the cloud and into our warehouses, stores, and eventually, our daily lives.

The 'Systems-First' Blueprint for an Embodied Future

At the heart of Striding AI’s strategy is a concept that diverges from the traditional, task-specific approach to robotics. The company is championing a "systems-first" methodology, focusing on what it calls Physical AI—a branch of intelligence that enables machines to interact with and learn from their physical environment. This is a crucial distinction from generative AI, which operates primarily in the digital realm of text and images. Physical AI is about embodiment: grounding digital intelligence in physical action.

To achieve this, the company is developing "robotic foundation systems." These are large, pre-trained AI models, analogous to the large language models (LLMs) that have revolutionized digital content creation, but built for physical interaction. They integrate vast datasets encompassing vision, language, and real-world actions to create a unified intelligence that can control a robot's hardware. This allows a robot to not just follow pre-programmed instructions but to generalize from its experience, understand complex commands, and adapt to novel situations—a capability that has long been the holy grail of robotics.

"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, in the company’s launch announcement.

This philosophy is a direct response to a fundamental challenge in the industry. For years, deploying robots in dynamic environments has been hampered by brittle software and the immense difficulty of programming for every possible contingency. By focusing on a foundational model that learns, Striding AI aims to create a scalable platform where every robot in the network contributes data to a central intelligence, creating a flywheel of continuous improvement. The goal is to build a system that learns from experience, much like a human, but at the collective scale of a fleet of machines.

A Strategic Beachhead in Retail

While the long-term vision is vast, Striding AI’s initial deployment strategy is remarkably pragmatic. The company is targeting structured retail environments for its first wave of robots, focusing on tasks like shelf restocking, inventory counting, and product organization. This choice is no accident. The global retail sector is grappling with persistent labor shortages, rising operational costs, and the critical need for accurate, real-time inventory data to compete with e-commerce giants.

According to market analysts, the retail robotics market is projected to grow exponentially, potentially exceeding $100 billion by the end of the decade. Companies like Brain Corp and PAL Robotics have already deployed robots for shelf-scanning and inventory audits. However, Striding AI’s approach promises a leap forward. Instead of single-task robots, its foundation systems are designed to enable machines that can handle a wider variety of tasks with greater adaptability. For example, a robot powered by a foundation model could potentially understand a command like "Tidy up the cereal aisle and make sure the new promotional items are front-facing," a task that requires complex perception, planning, and manipulation far beyond simple inventory scanning.

This initial focus on retail provides the perfect training ground. Supermarkets and warehouses are complex but repetitive environments, offering a rich stream of operational data and frequent, low-stakes human-robot interactions. These settings will serve as living laboratories for refining the company's core technology and proving its value in a demanding commercial landscape.

The Learning Flywheel: Data, Models, and Human-in-the-Loop

The technical core of Striding AI's promise lies in its infrastructure for learning. The company is building a platform for robot pre-training, distributed reinforcement learning, and edge-to-cloud orchestration. This isn't just about a single smart robot; it's about creating a networked intelligence.

A key component of this is a method called human-in-the-loop reinforcement learning (HIL-RL). In this paradigm, humans don't just program the robot; they act as teachers, providing feedback, demonstrating correct actions, and intervening to correct mistakes. This human guidance dramatically accelerates the learning process. Striding AI claims that in early internal tests, this method improved task success rates by up to three times. While an internal figure, this aligns with broader academic research showing that HIL-RL can significantly reduce the costly and time-consuming process of gathering real-world training data from scratch.

This data is then fed into a distributed learning system, where the experiences of every robot in the field are used to refine the central foundation model. An improvement made by one robot in a Beijing supermarket could, in theory, be propagated overnight to an entire fleet operating globally. This creates a powerful "flywheel effect"—the more robots deployed, the more data they collect; the more data they collect, the smarter the central model becomes; and the smarter the model, the more capable and valuable the robots are, driving further deployment.

This edge-to-cloud architecture is the invisible backbone that will support the company's vision. It’s a complex ballet of data processing, model training, and software updates that aims to make robots that not only work out of the box but get progressively better with every task they perform. The capabilities developed in the structured aisles of a retail store—handling diverse objects, navigating dynamic spaces, and executing complex plans—are designed to be transferable, forming the building blocks for more advanced applications in other industries. Through this systems-first approach, Striding AI aims to build robots that learn from the real world and gradually, almost imperceptibly, become a reliable part of our everyday environments.