Manufacturing
Mentee Robotics Pioneers Autonomous Humanoid Robot With NVIDIA Isaac
Mentee Robotics
Objective
Mentee Robotics is tapping into the NVIDIA Isaac™ platform—including its simulation and learning frameworks, and NVIDIA Jetson—to develop its humanoid robot, MenteeBot. The robotics company follows a simulation‑first approach, developing synthetic data generation pipelines in NVIDIA Isaac Sim™ and applying imitation and reinforcement learning techniques in NVIDIA Isaac Lab to train its policy models. These models are then deployed on MenteeBot, powered by Jetson, for real‑world operation.
Customer
Mentee Robotics
Use Case
Robotics
Products
NVIDIA Isaac
NVIDIA Jetson
NVIDIA Omniverse
Key Takeaways
- Multi-modal robot learning integrates vision, language, and motor control through the NVIDIA AI stack for complex behavior acquisition.
- Accelerated robot learning through simulation using Isaac Sim and Isaac Lab reduces risks, deployment costs, and the need for real-world data collection.
- Continuous learning from production variations reduces manufacturing costs and improves production yield.
- 25 kg payload capacity and hot-swappable batteries enable 24/7 learning-enhanced operations.
- An advanced robot learning pipeline using reinforcement learning-trained motion models can be deployed on dual NVIDIA® Jetson AGX Orin™ modules for real-time performance.
Accelerating MenteeBot Development With NVIDIA’s Three-Computer Solution
Mentee Robotics accelerates its intelligent robot learning system with the NVIDIA three-computer solution, which features dedicated platforms for training, simulation, and edge deployment.
The process begins on the NVIDIA DGX™ platform, where developers train and refine advanced AI models for humanoid perception, movement, and manipulation.
To post‑train its models through imitation learning, Mentee Robotics has built a custom synthetic data generation pipeline, developed entirely on Isaac Sim. The process starts with collecting expert human demonstrations using teleoperation. Demonstration data is collected by using either a motion planner or teleoperation in simulation. To efficiently scale existing trajectories in simulation, the team uses NVIDIA GR00T‑Mimic, with additional diversity introduced through Isaac Sim’s domain randomization—including randomized object positions, backgrounds, textures, and more.
Mentee Robotics also uses Isaac Lab to train policy models for motion control and dexterity using reinforcement learning. The resulting models are validated in Isaac Sim before running on a private cloud powered by NVIDIA RTX™ 6000 Ada GPUs.
The trained policy models are deployed on MenteeBot V3.0, equipped with dual NVIDIA Jetson AGX Orin processors, supporting autonomous and efficient processing for complex tasks such as visual perception, manipulation, and mobility. Looking ahead, Mentee Robotics plans to transition to NVIDIA Jetson Thor™ for even greater scalability and performance at the edge.
Accelerated Skill Acquisition Through Simulation
Simulation-based training through Isaac Sim and millions of virtual interactions allows robots to rapidly master new behaviors that transfer smoothly to real-world deployment.
As a result, Mentee Robotics is demonstrating measurable gains in task-execution speed, manipulation accuracy, and adaptability. Robots update their protocols to fit new equipment, workflows, and safety standards, minimizing downtime and deployment effort. The learning architecture is designed for scale, supporting consumer applications like household assistance and elderly care, and enables fleet-wide rollout of learned behaviors with tailored localization for each deployment. This positions Mentee Robotics at the forefront of embodied AI innovation.
Autonomous Operation in Complex Environments
MenteeBot is trained and validated on Mentee Robotics’ advanced robot learning platform and is now being deployed across diverse industry segments to unlock new levels of autonomy.
Industrial Manufacturing Learning Applications
In manufacturing, variations in spool size, weight, and positioning slow replacement tasks and often require human intervention. To solve this, a proof‑of‑concept demonstrated a robot that adapts its manipulation strategies using reinforcement learning. It continually refines performance, cutting task times and boosting precision through experience‑based optimization and handling variability with increasing autonomy over time.
Retail and Logistics Learning Adaptation
Collaboration with a major European retailer validated MenteeBot's ability to learn and adapt fulfillment center logistics behaviors, including learning optimal box-picking and cart-loading strategies, and adjusting material transport routes based on dynamic warehouse conditions. Menteebot adapts to inventory layouts, human worker patterns, and operational priorities in real time while maintaining safety and efficiency standards.
“NVIDIA's hardware and AI stack provided the compute efficiency, scalability, and real-time AI processing required for advanced robot learning and humanoid autonomy. Our comparisons showed that alternative solutions lacked the necessary AI-first architecture and high-performance hardware integration for continuous learning and adaptation.”
Prof. Lior Wolf
Mentee Robotics Co-Founder & CEO
Delivering Cutting-Edge, Adaptive Systems
Mentee Robotics has dramatically accelerated its humanoid robot development and reduced traditional development bottlenecks. The Isaac Sim simulation-first approach allowed the team to rapidly iterate on robot behaviors without the time and cost constraints of physical testing, while the custom synthetic data generation pipeline scaled up training data exponentially compared to manual collection methods.
The ability to validate models in simulation before physical deployment also minimized costly iterations, while Isaac Lab's reinforcement learning capabilities enabled continuous improvement cycles that are typically tedious and time-consuming to achieve through traditional robot learning methods.
Mentee Robotics has positioned itself at the forefront of embodied AI innovation, delivering learning-capable humanoid robots that continuously adapt and improve in real-world environments.
Discover how the NVIDIA robot learning platform—including Isaac GR00T, Isaac Lab, and Isaac Sim—can help you train adaptive, intelligent robots faster and more cost-effectively.