The Internet can't stop watching Figure AI's humanoid robots handling packages

2026-05-20 · Source: AI - Ars Technica · Field: Technology & Digital — Robotics & Autonomous Systems, Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Intermediate, short

Summary

Figure AI recently livestreamed its Figure 03 humanoid robots performing a package handling task for nearly a week, starting May 13. The robots autonomously inspected barcodes on various packages and placed them on a conveyor belt, demonstrating "long horizon autonomy" powered by the onboard Helix 02 neural network, trained on over 1,000 hours of human motion data and 200,000 simulation environments. This viral demonstration, initially planned for eight hours, extended to 24/7 operation, attracting significant online attention and even Polymarket bets. A "Man vs. Machine" competition on May 17 saw Figure AI intern Aimé Gérard sort 12,924 packages (2.79 seconds/package) against the robots' 12,732 packages (2.83 seconds/package), with the human winning. While impressive, the demo highlights the robots' current limitations to specific tasks and environments, despite Figure AI's broader vision for general-purpose humanoid robots and its nearly \$2 billion in funding.

Key takeaway

For Robotics Engineers evaluating humanoid robot deployments, recognize that even extensive public demonstrations like Figure AI's livestream offer limited insights into general-purpose capabilities. You should prioritize independent verification of autonomy claims and assess performance across diverse, unstructured environments. Focus on long-term reliability metrics beyond initial endurance tests, as current humanoids still lag human efficiency in specific tasks.

Key insights

Humanoid robot demonstrations, even impressive ones, offer narrow views of real-world capabilities and often face scrutiny regarding true autonomy.

Principles

• Long-duration robot demos build trust.
• Onboard AI inference enables autonomy.
• Human motion data improves robot control.

Method

Figure AI's Helix 02 system uses a whole-body controller trained on >1,000 hours of human motion data and >200,000 parallel simulation environments for full-body control and long horizon autonomy.

In practice

• Network robots for battery swaps.
• Use livestreaming for transparency.
• Benchmark against human performance.

Topics

• Humanoid Robots
• Figure AI
• Robot Autonomy
• Warehouse Automation
• AI Inference
• Human-Robot Competition

Best for: Investor, Robotics Engineer, AI Engineer, Tech Journalist

Related on AIssential

• AI Glasses With an Agent Inside — Figure's humanoid robots combine robust hardware with advanced AI for autonomous, general-purpose physical tasks.
• The skeptic’s guide to humanoid robots going viral on the Internet — Viral robot demos often mislead by implying human-like generalization and autonomy not yet achieved in real-world applications.
• Humanoid Robots Are Coming to Factory Floors — But the Hardware Isn’t Ready Yet — Humanoid robot hardware, not AI, is the critical bottleneck for industrial deployment due to environmental demands.
• Physical AI Matters More Than Humanoid Robots — Physical AI integrates perception, reasoning, and action to enable AI systems to interact effectively with the real world.
• Perplexity - Perplexity — Humanoid robots are nearing human-level performance in repetitive warehouse tasks, signaling rapid automation advancements.
• The Download: gig workers training humanoids, and better AI benchmarks — AI training and evaluation methods require significant evolution to meet real-world demands and ethical considerations.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by AI - Ars Technica.