Mentee Foundation Model

Mentee Foundation Model

The Mentee Foundation Model is Mentee Robotics' in-house AI platform and the learning and control core of the MenteeBot humanoid. Unlike approaches based on external frameworks (ROS, MoveIt), Mentee built its own fully vertically integrated software stack.

Learning method

The system operates in two main learning modes. "Learning from data" covers language inputs (LLM), vision (NeRF-based 3D mapping), and speech, allowing the robot to build a cognitive map of the world. "Learning from experience" is based on reinforcement learning (RL) in NVIDIA Isaac Sim with Sim2Real policy transfer to the physical robot.

Human mentoring

A key feature of the platform is learning from demonstration: a human mentor shows the robot a new task in real time, and the robot learns to reproduce and execute it autonomously. This paradigm — analogous to on-the-job training for a new employee — enables rapid acquisition of new skills without code changes.

Tech stack

The Foundation Model runs on two NVIDIA Jetson AGX Orin modules installed directly on MenteeBot V3, providing redundancy and on-board heavy AI workload capability. Simulation and RL training are performed in NVIDIA Isaac Sim and Isaac Lab.