Industrial robots are incredibly good at doing the same thing over and over again.
Put a car door in exactly the right place, and a robot can weld it thousands of times without getting tired or distracted.
But the real world rarely cooperates like that.
Objects move. Lighting changes. A box gets turned sideways. Or something winds up in a place the robot has never seen before.
Humans can adjust to those kinds of changing conditions without even thinking about it. But robots usually struggle to adapt on the fly.
Today, artificial intelligence is helping them overcome that limitation.
And this week’s chart shows just how quickly robots are learning to deal with a world that refuses to stay put.
Teaching Robots to Adapt
To understand what’s happening, we can look at a robotics test called SIMPLER.
The idea behind it is pretty simple. Researchers give a robot a task, like picking up an object and putting it somewhere else. Then they see whether the AI controlling the robot can get the job done.
But they don’t make it easy.
Researchers can change the lighting, background or even the texture of the table. They can also add distracting objects to the scene.
That means the robot can’t simply memorize what everything is supposed to look like.
It has to adapt.
And that gives researchers a way to measure something that could be incredibly important for the future of physical labor: How well can an AI-powered robot deal with changing conditions?
Take a look at this week’s chart.
Before I get into the numbers, I want to be clear about how we created this chart.
It wasn’t published by SIMPLER. We built it using results reported by researchers who tested three different robot AI models on the same SIMPLER Google Robot benchmark.
Specifically, we’re looking at their average success rates across three tasks: picking up a Coke can, moving one object next to another and opening or closing a drawer.
And as you can see, the improvement over the past two years is significant.
OpenVLA, an open-source Vision-Language-Action model released in 2024, successfully completed an average of just under 40% of these tests.
A newer model, X-VLA, pushed that number to nearly 76%. And the newest model on our chart, TBD-VLA, reached more than 86%.
That means the success rate on these particular tests has more than doubled in roughly two years.
And some of the individual results are even more impressive.
TBD-VLA successfully picked up the Coke can 97% of the time. And it completed the drawer tests more than 83% of the time.
To be clear, these results don’t mean that robots can successfully perform 86% of the jobs humans do.
Not even close.
SIMPLER is a simulation. And the three tasks represented here are only a tiny fraction of what a human worker might encounter during a day.
But these results are encouraging when you think about what happens inside a warehouse or factory.
Boxes don’t always face the same direction. Parts can get moved around, lighting can change or something can get knocked over and wind up where it isn’t supposed to be.
A traditional industrial robot works best when engineers remove as much of that uncertainty as possible. That’s why factories have historically built entire work areas around them, so the machine knows exactly where the next part will arrive and exactly what movement it needs to make.
But AI-powered robots can work differently.
Instead of programming every movement ahead of time, engineers can give the robot an instruction and use AI to help it understand what it sees and determine how to complete the task.
And as this week’s chart shows, AI systems are getting much better at doing that.
Here’s My Take
Robots still have a way to go before they can handle the unpredictable physical world as easily as we can.
But this week’s chart shows how quickly that gap is narrowing.
As robots become better at adapting to their surroundings, they’ll become useful in far more places than a carefully controlled factory floor.
And as I wrote about last week, that could make them even more valuable at a time when America struggles to find enough workers.
Regards,
Ian King
Chief Strategist, Banyan Hill Publishing
Editor’s Note: We’d love to hear from you!
If you want to share your thoughts or suggestions about the Daily Disruptor, or if there are any specific topics you’d like us to cover, just send an email to dailydisruptor@banyanhill.com.
Don’t worry, we won’t reveal your full name in the event we publish a response. So feel free to comment away!
