Home AI Scientists slash robot setup time from months to just two hours

Scientists slash robot setup time from months to just two hours

Researchers set up and operate robotic arms using RIO’s unified interface for robot control and teleoperation. Credit: Carnegie Mellon University.

Teaching a robot to perform new tasks is often much harder than people imagine.

Before researchers can even begin testing artificial intelligence (AI), they may spend weeks or even months simply setting up the robot’s software.

Now, scientists at Carnegie Mellon University have developed a new open-source software framework that could dramatically speed up that process.

The new system, called Robot I/O (RIO), is designed to make it much easier to use AI on different kinds of robots.

Instead of creating new software every time researchers switch to a different machine, RIO provides a common foundation that works across many robotic platforms.

The project aims to solve one of the biggest problems in robotics today. While AI has advanced rapidly, the software infrastructure needed to support robots has not kept pace.

Many research groups build custom software for each robot they own. As a result, programs created for one robot often cannot be used on another.

Researchers frequently have to rewrite large amounts of code before they can start a new project, slowing progress and making it difficult for different laboratories to share their work.

RIO was designed to remove much of this extra effort.

The software provides a single interface for controlling robots, collecting data, operating robots remotely and running AI systems. It works with different types of machines, including robotic arms, humanoid robots and other robotic platforms.

Instead of rebuilding everything from the beginning, researchers can reuse software components and simply swap in the parts they need for a particular robot. This makes the entire development process much faster and more flexible.

The benefits became clear during testing. An undergraduate student with experience in machine learning but no background in robotics was asked to unpack a robotic arm and set it up using RIO.

Following the project’s instructions, she successfully went from opening the box to controlling the robot in about two hours. Without a system like RIO, this process could normally take far longer for someone new to robotics.

The researchers believe this easier setup could help students begin meaningful research much sooner instead of spending months learning complicated robot software.

Another major advantage is improved collaboration. Because many robotics laboratories currently use different software systems, sharing AI models, data and experiments can be difficult.

RIO allows researchers to use the same software pipeline across many different robots. Whether a robot has extra cameras, additional arms or a different body design, much of the software can remain the same.

This also makes scientific research easier to reproduce because other laboratories can use the same tools instead of rebuilding them from scratch.

The team says the need for this type of shared infrastructure has become even more important as powerful AI systems continue to improve. Modern robots require huge amounts of training data, but collecting that data depends on having reliable software that can manage many different robots efficiently.

Although RIO is still under active development, some members of the research team are already expanding the technology through a startup company called Lavoro AI. Their long-term goal is to make deploying robots much simpler and help create AI systems that can quickly adapt to new tasks, environments and hardware.

By reducing the technical barriers that slow robotics research, RIO could help scientists spend less time writing setup code and more time developing smarter, more capable robots.

In the future, that could speed up advances in manufacturing, healthcare, home assistance and many other fields where intelligent robots are expected to play an increasingly important role.