The aftermath of the World Humanoid Robot Games has brought bustling activity to Unix AI’s customer hotline, with a surge in inquiries and visits from interested parties. Yang Fengyu, the CEO of Unix AI, shared that following their remarkable success by winning gold and silver medals at the international event in the hotel cleaning and guest reception categories, they attracted attention from various sectors like hotels, retirement homes, and service operators venturing into service robotics.
The competitions focused on testing robots’ capabilities in generalization, dexterity, and speed. Tasks included cleaning rooms swiftly and efficiently, and assisting guests by retrieving and transporting their luggage. Unix AI’s robots showcased exceptional performance, a testament to their extensive experience and deployment in simulated consumer environments, particularly hotels. These robots have been effectively handling cleaning tasks within hotels, learning on the job, and collecting valuable data.
Yang envisions the skills acquired in hotel settings to be transferable to households, restaurants, cafes, and fast-food establishments in the future. Unix AI’s approach notably differs from other robotics startups, as they opt for a method that breaks down tasks into key points and motion trajectories, utilizing imitation learning rather than the end-to-end VLA approach due to limited training data. This strategy has enabled their robots to quickly learn new movements with minimal demonstrations and continuously enhance their performance through practical deployment.
Yang Fengyu, a young entrepreneur with a background in computer science, recognized the opportunity to establish Unix AI leveraging China’s supply chain and market advantages. He highlighted the pivotal role of real-world deployment in scaling robotics, emphasizing the importance of data diversity and scale for building robust models. Unix AI’s technology features a perception-operation decoupled model based on key-point imitation learning, aiming to enhance robots’ adaptability to various tasks.
The company’s innovative UniTouch system integrates vision and tactile data to improve material recognition and feedback, enhancing robots’ handling capabilities. While challenges exist in developing tactile sensing due to cost and durability concerns, Unix AI emphasizes the significance of hardware stability in the era of mass production of robotics. By prioritizing in-house development for hardware and maintaining control over the supply chain, they ensure data consistency and reduce costs, ultimately benefiting their product pricing and performance.
The upcoming third generation of Unix AI’s humanoid robot, Wanda, is designed with a focus on functionality rather than human likeness, boasting enhanced features such as omnidirectional wheels, increased load capacity, and improved height control. Yang envisions a future where youthful energy and technical expertise drive progress in embodied intelligence, fostering a dynamic synergy with seasoned industry veterans. Ultimately, he believes that combining fresh perspectives with established industry know-how yields a potent blend propelling technological advancements forward.
About the Author
