Mentee’s goal is to fully integrate robotics, sensing, and AI into a humanoid robot with human-like dexterity and perception, and sufficient reasoning capabilities to execute a diverse array of tasks and continuously adapt and learn new ones.

While the field of humanoid robotics is booming and new robots are being unveiled every month, the gait of these robots remains stiff and closely resembles the way Honda’s ASIMO walked two decades ago.

Mentee Robotics employs deep reinforcement learning to develop control policies for its range of humanoid robots.

The company focuses on technologies that minimize the discrepancy between simulation and the real world and close the loop by mapping the real world back to the simulation.

As demonstrated, unlike classical control methods, the resulting locomotion allows the robot to walk forward and backward, run, and turn efficiently, exhibiting agility and human-like movement.

Similarly, when applied to arm dexterity and grasping, RL-based control results in a natural motion that differs from that obtained with inverse kinematics.

The ability to swiftly train control policies for new or altered mechanics is revolutionizing the development cycles in robotics, allowing for rapid evolution of both the mechanical design and the software.