This Humanoid Robot Is a Terrifyingly Competent Office Intern

2026-06-29 · Source: WIRED - Ai · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning · Depth: Intermediate, short

Summary

Flexion Robotics, a Swiss startup founded by ex-Nvidia researchers, has developed a novel method for training humanoid robots to perform complex tasks. Their approach involves teaching individual skills in simulation, then using a master AI algorithm to combine these skills for real-world execution, moving beyond traditional teleoperation. A modified Unitree robot demonstrated this by autonomously retrieving, unpacking, and placing snacks, navigating stairs and an elevator. The core of Flexion's software learns actions from human videos and triggers pre-learned skills, with extensive reinforcement learning applied across all AI layers. While industry leaders foresee humanoids significantly impacting the economy, analysts emphasize that the AI models, not the hardware, are the truly transformative element. ABI Research projects the robot foundation model market could reach \$150 billion by 2036. An embedded video also highlighted a recent half marathon in China where autonomous humanoid robots, like Honor's "Lightning," competed and set new records, despite some public setbacks, underscoring the ongoing US-China race in humanoid development.

Key takeaway

For Robotics Engineers developing autonomous systems, Flexion's approach highlights the critical role of AI models over hardware. You should prioritize robust simulation environments for skill acquisition and integrate reinforcement learning across your software stack. Focus on systems that learn from human demonstrations and can adapt skills to new contexts, as this capability is crucial for commercial viability and overcoming the limitations of teleoperation in complex, unfamiliar settings.

Key insights

Flexion Robotics trains humanoids by combining simulated skill learning with a master AI interpreting human video actions.

Principles

• AI models are the core of humanoid utility.
• Simulation-based skill learning enhances autonomy.
• Reinforcement learning optimizes task mastery.

Method

Flexion's method involves digesting human action videos to inform a master AI, which then triggers individual skills learned via reinforcement learning in simulation for real-world execution.

In practice

• Train specific skills in simulation.
• Use human video for action sequencing.
• Apply reinforcement learning broadly.

Topics

• Humanoid Robotics
• AI Training
• Reinforcement Learning
• Robot Foundation Models
• Simulation Learning
• Autonomous Systems

Best for: Research Scientist, Investor, Robotics Engineer, AI Scientist, Director of AI/ML

Related on AIssential

• Why Do Humanoid Robots Still Struggle With the Small Stuff? — Despite AI advancements, humanoid robots still struggle with fundamental physics like force control for reliable real-world interaction.
• Video Friday: Humanoid Learns Tennis Skills Playing Humans — Robotics research is advancing dexterous manipulation, sustainable design, and complex locomotion through AI and novel hardware.
• Video Friday: Digit Learns to Dance—Virtually Overnight — Advanced robotics are achieving new performance thresholds in control, learning, and human interaction.
• Operating a Humanoid With Your Body Is a Hot Job in China’s Hardware Capital — Human teleoperation of humanoid robots collects training data for eventual autonomous functionality in workplaces.
• AI Glasses With an Agent Inside — Figure's humanoid robots combine robust hardware with advanced AI for autonomous, general-purpose physical tasks.
• TaskNPoint: How to Teach Your Humanoid to Hit a Backhand in Minutes — Dynamic robot skills are mastered by focusing on crucial interaction windows and practicing distinct actions.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by WIRED - Ai.