Genesis AI, a French robotics startup with an R&D center in Silicon Valley, announced a significant breakthrough in robotic manipulation on Wednesday. Backed by Eclipse and former Google CEO Eric Schmidt, the company demonstrated that its robots can perform complex tasks like playing the piano and cracking eggs at speeds and precision approaching human levels. The demonstrations, shown as real-time videos without teleoperation, highlight the startup's progress in solving one of the most difficult challenges in robotics: enabling machines to grab, move, and assemble objects with dexterity. During the event, a video showcased robotic hands keeping pace with a fast-paced piano composition of approximately 130 beats per minute. The team also displayed a robot cracking an egg with a single hand and manipulating wires. CEO Zhou Xian clarified that these demos were executed autonomously at 1x speed. However, he noted the robots are not yet capable of zero-shot execution, meaning they require specific training for individual tasks. For instance, teaching the robot a new piano piece took the 60-person team about one hour, while the egg-cracking demo required a few hundred recorded human trajectories to train the system effectively. Regarding performance metrics, the robot achieved a 90% to 95% success rate for most cooking steps, though simpler subtasks like one-handed egg cracking hovered around 50% to 60%. Xian stated the robot operates at roughly 60% to 70% of human speed and considers these feats among the most complex ever performed by a robot in a human-like manner. Unlike competitors who focus solely on AI models, Genesis AI is a full-stack company developing its own AI models, robot hardware, training gloves, and simulators. The startup aims to create a general-purpose robot capable of functioning in both factories and homes within the next decade. A key differentiator is the robot hand, which features 20 degrees of freedom and 20 motors integrated directly into the fingers, rather than relying on external cables or tendons. This design allows for finer control similar to human anatomy. To gather data, Genesis uses proprietary training gloves that capture motion and tactile force signals, supplementing public internet data. The company is also exploring partnerships with industrial clients to have employees wear these gloves during work to collect real-world datasets. Additionally, an in-house simulator allows the team to test models in virtual environments before deploying them on physical hardware, significantly accelerating the development cycle. Xian emphasized that while the company has not yet solved manipulation entirely, this approach represents a critical step forward. He expressed a commitment to pushing the field beyond the status quo, arguing that the future lies in building robots that are fundamentally indistinguishable in capability from human workers. The company remains ambitious, targeting a future where the distinction between industrial and home robots vanishes as they evolve into versatile tools for diverse environments.