Why can ChatGPT draft an essay, yet struggle with folding laundry? This question drives researchers as they develop software enabling robots to comprehend and execute commands in the physical world. The pursuit of artificial intelligence in robotics is gaining momentum, aiming to bridge the gap between virtual capabilities and real-world applications.

As Ailsa Chang noted, artificial intelligence excels at tasks like finding recipes⁣ or generating images. However, the ability to physically “hang that picture on a wall or cook you dinner” remains a challenge. Researchers are actively working to integrate AI into robots, striving⁣ to make them functional in practical scenarios.

Stanford’s AI Robot: A Step⁢ Towards Real-World AI

At Stanford University, Moo jin Kim, a graduate student, is developing a novel robot ⁣powered by AI, drawing inspiration from the AI used in chatbots. Kim describes the project as:

It’s one step in the direction of, like, ChatGPT for robotics, but still⁢ a lot of work to do.

This robot, while physically simple ⁢with its mechanical arms and pincers, stands out due to its internal workings. Unlike conventional robots ⁣that require meticulous programming, this one uses a teachable AI neural network. Kim trains the robot by repeatedly demonstrating tasks:

So basically, like, whatever task you want to do, you just keep doing it over and over, maybe like, 50 times or 100 times.

Through this method, the robot’s AI neural network learns to perform tasks independently.

The Vision⁢ of Intelligent Robots

Chelsea Finn, who leads the laboratory at Stanford, envisions a future where robots can understand ⁣and execute commands on the fly. Her goal is to:

develop software that would allow the robots to operate intelligently in any situation.

By “intelligently,” she means the robot should be able to understand and act upon ⁢simple instructions in real-world contexts. Finn illustrates this with examples of basic tasks:

Even just to⁣ do very basic things, like being able to make a sandwich or being able to clean a kitchen or being able to restock grocery store shelves.

Finn is also the cofounder of Physical Intelligence, a startup ⁤that recently showcased a mobile robot capable of folding laundry, trained via human-guided AI programming. Finn explained:

In⁢ that case, we actually had a workstation that was in the apartment that ‍was computing the actions and then sending it over the network to the robot.

Challenges and Limitations of ⁢AI Robots

Despite these advancements, AI robots still face challenges. They can become confused, misinterpret instructions, make errors, or simply get stuck. Ken Goldberg,a professor at the University of California at Berkeley,emphasizes:

Robots are not going to suddenly become the science‍ fiction dream overnight.

Goldberg points out that while⁣ AI text-writing has considerably improved due to vast‍ amounts of training data, robotics lacks a similar foundation:

For robotics, there’s ⁢nothing. We don’t have anything to start with, right? There’s no examples online of robot commands being generated in⁤ response to robot inputs.

he estimates that at the current rate of data acquisition,it would take an impractical amount of time to gather sufficient data for robots to‍ learn effectively:

You know,at this current rate,we’re going to take 100,000 years to get that much data.

Exploring Alternative Training Methods

One potential solution involves training AI within computer simulations. researchers in Switzerland successfully trained a drone ‍to race by using a simulator. This allowed the drone to outperform human opponents in a real-world indoor course. However, simulations have⁢ limitations, as anything not simulated, such as weather conditions, can cause the drone to crash. Furthermore,simulating complex interactions like ‍folding laundry remains challenging.

The Core Problem: Framing the Task

Matthew Johnson-Roberson at Carnegie Mellon University suggests that the issue goes beyond the amount of data available. He argues:

In my mind, the question ⁣is not, do we have enough data? It is ‍more, what is the framing of the problem?

Johnson-Roberson explains that while AI chatbots ‍excel at predicting the next word in a sequence, ⁢robots⁢ must ⁢perform far more ⁢complex tasks. He questions ⁣the effectiveness of simply feeding robots large amounts of visual data:

Next best word prediction works really well, and⁤ it’s a very simple problem ’cause you’re just predicting the next word. And it is⁤ not clear right now – I can take 20 hours of GoPro footage and then⁣ produce anything ⁤sensible with respect to⁤ how a robot moves around in the ‍world.

Therefore,⁢ researchers need to develop better methods for teaching robots to handle the complexities of real-world tasks.

A Glimpse of Progress

Back at Chelsea Finn’s lab, Moo Jin Kim and an observer watched as the robot attempted to scoop trail mix. The robot⁣ successfully identified the correct bin, ‍a moment of relief for Kim, who noted:

Usually, that spot right there, where it⁣ identifies the object and goes to it, that’s the part where we hold our breath (laughter).

Despite a small scoop, it was a step ⁤in‍ the ⁢right direction, demonstrating the potential of AI-powered robots.