Attention, Boston Dynamics! Mentee Robotics Launches Next-Gen ‘AI-First’ Robot

Today, Mentee Robotics, a startup co-founded by Mobileye and AI21 Labs founder Amnon Shashua, has emerged from stealth mode to unveil its latest innovation: Menteebot, an AI-driven humanoid robot.

Still in the prototype stage, Menteebot is designed for household and warehouse applications, leveraging advanced AI technologies, including transformer-based large language models (LLMs) popularized by OpenAI's ChatGPT. This approach allows the robot to operate effectively across all layers of its functionality.

The Next Generation of AI Agents

Menteebot is positioned as an AI-first robot capable of completing complex tasks from start to finish. Unlike many competitors that integrate AI into existing systems, Mentee Robotics built Menteebot from the ground up with an AI-centric design philosophy.

The company released a video showcasing Menteebot responding to verbal commands, demonstrating its ability to process tasks while exhibiting locomotion, scene understanding, object detection, and grasping capabilities.

What Makes Menteebot Unique?

While humanoid robots have been developed for years, most have focused on physical interactions, such as mimicking human movement. Previous generations of robots were typically pre-programmed or operated via software for specific tasks, like sorting items in controlled environments.

Recent advancements in language and embodied learning models have enabled a new wave of robots to understand and respond to natural language commands. Mentee Robotics distinguishes itself by integrating AI throughout Menteebot's entire operational framework rather than retrofitting existing designs.

The Three Levels of Menteebot

Menteebot converts human commands into complex actions using AI at three main levels:

1. Command Interpretation: Utilizing transformer-based LLMs, Menteebot understands commands and strategizes the steps necessary to complete tasks.

2. Environmental Mapping: Through NeRF-based algorithms, it creates a cognitive 3D map of its surroundings, gathering semantic information about objects and localizing itself to navigate obstacles dynamically.

3. Execution of Tasks: Employing a simulator-to-reality (Sim2Real) machine-learning approach, Menteebot executes its plan with authentic gait and hand movements adapted from a simulated environment.

Amnon Shashua noted, “We are on the cusp of a convergence of computer vision, natural language understanding, and sophisticated simulators. At Mentee Robotics, this marks the foundation for designing a general-purpose bipedal robot capable of performing household tasks and learning through imitation.”

The demo video showcases Menteebot moving fruit from one location to another in a kitchen, although it requires multiple commands to perform the task rather than executing it in one fluid motion. As a prototype, continuous improvements are expected, ultimately enabling Menteebot to handle complex commands seamlessly.

Mentee anticipates that the final production model will feature camera-only sensing, proprietary electric motors for enhanced dexterity, and fully integrated AI, with a projected launch in the first quarter of 2025. However, the targeted market segment remains unconfirmed.

Competition in AI-Driven Humanoids

While Mentee Robotics benefits from Shashua’s expertise in computer vision and LLMs, it faces stiff competition. Companies like Tesla, OpenAI-backed Figure AI, and 1X Technologies are actively developing AI-powered humanoids. Nvidia has also introduced Project GR00T, a foundational model for humanoid robots, partnering with various industry players, including Agility Robotics, Boston Dynamics, and Unitree Robotics.

Boston Dynamics has recently revealed the all-electric Atlas humanoid robot, aimed at automotive and industrial applications. The competitive landscape promises to be dynamic as Mentee Robotics works to deploy its AI-first humanoid in this rapidly evolving field.