Carnegie Mellon University researchers have introduced an open-source software framework designed to help AI systems move more easily between different robots, tackling a persistent bottleneck that can delay robotics research before experiments even begin.
The framework, called Robot I/O, or RIO, gives researchers a shared interface for robot control, data collection, teleoperation and AI deployment, CMU said. The goal is to reduce the custom engineering work often required when teams move from one robotic arm, humanoid or bimanual platform to another.
The infrastructure gap holding back robot learning
For many robotics teams, the hard part is not always the idea but getting the machine ready, as researchers can spend weeks or months configuring a new robot before testing any behavior, rewriting software and rebuilding data pipelines that may already exist in another form for a different system.
Jean Oh, an associate research professor in CMU’s Robotics Institute, said infrastructure has become one of the biggest bottlenecks in robot learning research, and RIO is designed to make that first step faster.
In one test, undergraduate intern Reya Shukla, who had machine learning experience but no robotics background, unpacked a robotic arm and began controlling it through RIO in about two hours, according to CMU.
Reusable tools could speed up robot learning
The need for shared infrastructure is growing as robotics moves toward more general-purpose AI systems, where progress depends on more data, more testing and reusable tools that can work across labs and machines.
However, that progress still depends on the physical world, because robot data has to be collected from real machines, which means researchers need reliable infrastructure before they can train stronger models.
RIO’s modular design lets teams reuse building blocks instead of writing special code for every robot, sensor setup or policy-training pipeline, making experiments easier to reproduce and helping researchers build on each other’s work instead of starting from scratch.
From lab prototypes to real-world machines
Beyond experimentation, the framework could also help move robot learning closer to real-world deployment.
According to Bosch’s Jonathan Francis, industrial robotics rarely depends on one robot, one sensor setup or one fixed environment, meaning RIO could make it easier to reuse, test and adapt robot learning systems across different platforms outside the lab.
CMU said RIO remains an active research project, with future work focused on expanding hardware support and making new robots easier to bring online, while some team members are building on the technology through Lavoro AI, a startup co-founded by Oh.
