Figure Tested Its New Robot in 30 Homes. Did It Work

Figure has finally taken its humanoid robot out of the lab and tested it across 30 unfamiliar homes. The robot picked up toys, folded towels, and made beds — successfully completing 56% of its attempts.
So is 56% impressive… or a failure?

In this video, we break down what Figure’s results actually mean, why comparing its 56% success rate directly with a competitor’s 99% is misleading, and how pretraining on millions of videos of humans performing everyday tasks boosted the robot’s performance from roughly 8% to 56%.

Figure is now paying tens of thousands of people to record themselves cleaning, cooking, doing laundry, and performing other everyday tasks. Tesla, DoorDash, and companies in China are pursuing similar strategies. The robotics industry is making a multibillion-dollar bet that robots can learn the physical world the same way large AI models learned language: with massive amounts of data and compute.
But there is a problem. Human video contains no sense of touch, and impressive demos are still a long way from a reliable robot that can work safely and independently in your home.

We examine Figure’s 30-home experiment, its massive Index dataset, the global race for human video, the billions being spent on compute, criticism of the approach — and when you might actually be able to buy a household robot that works.

00:00 Introduction
00:44 A Robot Cleans 30 Real Homes
04:39 Is a 56% Success Rate Good or Bad?
06:40 56% vs. 99%: Comparing Robot Success Rates
09:42 Index: How Figure Trains Its Robots
12:31 The Race for Human Video — and Billions in Compute
16:23 Criticism, Safety, and Trust
20:00 What Can You Buy Today — and When Will the Perfect Home Robot Arrive?

#machine #Robotics #robot