The Menteebot prototype unveiled today incorporates artificial intelligence at all levels of its operation. The innovative approach uses machine learning from simulation to reality (Sim2Real), allowing training in a simulated version of the robot with unlimited data before adapting to the real world with minimal data requirements. Real-time environment mapping is done using NeRF algorithms, creating cognitive maps for the robot to navigate and avoid obstacles.
Large transformation-based language models interpret commands and coordinate the robot’s movements for complex tasks. The prototype demonstrates end-to-end action, from verbal commands to completing tasks like navigation, understanding the environment, identifying objects, and grasping. The robot is expected to go into serial production in 2025, equipped with camera-based sensing, electric motors for agility, and fully integrated artificial intelligence for complex thought processes.
The company’s founder, Prof. Amnon Shashua, sees a convergence of technologies in AI, computer vision, natural language understanding, and simulation-to-real-world methodologies. The robot’s AI capabilities will enable it to perform household tasks, learn new tasks through imitation, and move freely in human environments. This marks the beginning of designing walking human robots for various purposes.