Ex-Nvidia Engineers Build AI System Training Humanoid Robots for Office Tasks

Executive Briefing

• Flexion Robotics trains humanoid robots using reinforcement learning in simulation, reducing reliance on human teleoperation
• Demonstrates robot autonomously navigating stairs, elevators, and unpacking parcels from a single natural-language command
• Combines multiple AI layers—master model, simulation, and motor control—each using trial-and-error reinforcement learning
• ABI Research projects the robot foundation model market could reach $150 billion by 2036
• Flexion's software works across multiple humanoid hardware platforms, potentially increasing its commercial appeal