An autonomous robot walks through the corridors of the Sherman Fairchild Library as if searching for a book. Legs bend, steps measured, no rush. It reaches the open air, stops at a precise point, and leans forward as if tying its shoelaces. Something detaches from its back: a drone. It takes off, flies over the campus pond, lands on the other side, and then something strange happens. It doesn't fold up, it doesn't turn back. It extends four wheels and begins to drive. I thought I'd seen it all. And now this pops up.
Caltech and the Technology Innovation Institute of Abu Dhabi They spent three years building this system. It's called X1, and it is the first autonomous robot that combines three ways of moving without asking anything from anyone.
How the autonomous robot that does three things (well) works
X1 isn't a single autonomous robot. It's a system. Two machines working together, each with its own task. The first is a Unitree G1 Modified: A humanoid about five feet tall, weighing 32 pounds, that walks using algorithms based on the physics of motion. It does not imitate recorded human footsteps. Learn to walk as gravity, balance, and the terrain dictate.. Can climb stairs, walk through corridors, carry loads on his back.
The second is M4, a transformable drone developed in the laboratories of Caltech. It can fly like a quadcopter and then, once landed, transform into a four-wheeled vehicle. Two modes, one machineThe idea is simple: flying uses battery power, driving saves it. The autonomous robot makes decisions based on what's in front of it.
The demo that explains why it is needed
The test was conducted on October 14, 2025, at the Pasadena campus. The scenario was simulated: an emergency is underway, and autonomous agents need to be sent to the scene. The autonomous robot departs from Gates-Thomas Laboratory, crosses the library, goes outside. He reaches a high point where he can safely release the M4. He folds. The drone releases, takes off in flight mode. It hovers over an area, lands, switches to driving mode, and continues on wheels.
At one point he meets the Turtle Pond, the campus pond. The wheels are no longer needed. It turns back into a drone, crosses the water, and reaches its destination. No remote pilot, no human commandOnly sensors, algorithms, and autonomous decisions.
Aaron Ames, Director of Center for Autonomous Systems and Technologies (CAST) from Caltech, explains the key point: "The robot learns to walk as physics dictates. It can move on different types of terrain, go up and down stairs, and most importantly, it can do so with things like the M4 on its back."It's not a remote control. It's true autonomy. The autonomous robot calculates, adapts, and continues.
Why it took three years to build an autonomous robot with a flying backpack?
The project was born from the collaboration between Caltech and the Technology Innovation Institute (TII) in Abu Dhabi. Three years of work, two continents, diverse expertise. The Ames lab at Caltech studied bipedal locomotion. The group Mory Gharib, also at Caltech, developed M4. TII brought Saluki, a secure flight controller and an on-board computing system designed to withstand cyber threats. La Northeastern University, with Alireza Ramezani's team, contributed to the transformable designEach piece had a role. The result was a system that didn't exist before.
The real problem wasn't building an autonomous walking robot, or a flying drone. Those have been around for years now.The problem was getting them to work together as a team. “The challenge is to make different robots collaborate so that they become a single system that offers different functionalities.”, says Gharib. With this collaboration, they found the right fit.
Where can such an autonomous robot be useful?
Earthquakes. Floods. Fires. Collapses. Situations where sending people is risky or impossible. Autonomous robots are already changing the way emergencies are handledDrones mapping affected areas, ground robots searching for missing people under rubble, autonomous vehicles transporting supplies.
X1 adds a new layer: the ability to adapt to the environment. Need to walk through rubble? Walk. Need to fly over a flooded area? Fly. Need to navigate an open road? Drive. One autonomous robot, three strategies.
Claudio Tortorici, director of TII, underlines the importance of perceptual autonomy:
"We install different types of sensors: lidar, cameras, rangefinders. We combine all this data to understand where the robot is, and the robot understands where it is to get from one point to another.".
The future goal is to equip the entire system with model-based algorithms and machine learning-driven autonomy to navigate and adapt to the environment in real time. Robotics in 2025 is aiming for precisely this: intelligent and adaptive systems..
What's still missing?
X1 works, but it's not yet ready for real-world operations. It lacks complete decision-making autonomy. For now, the system can execute a sequence of programmed actions: start here, walk there, release the drone, fly to that point. The next step is to make it capable of making strategic decisions without human supervision.
“We believe we are at a stage where people are starting to accept these robots.”, says Tortorici. “But for there to be robots everywhere around us, these robots must be reliable.”Ames confirms: the ongoing work concerns critical security controls, system trust, and cybersecurity. These are big problems that require multiple projects and collaborations.
The Caltech-TII collaboration doesn't stop with X1. Other projects are in the pipeline that explore different facets of robotic autonomy. "We have multiple projects that extend beyond this, studying all these different facets of autonomy. These are really big problems. By having these different projects and aspects of our collaboration, we're able to tackle much bigger problems and advance autonomy in a substantial and coordinated way.", explains Ames.
X1 isn't the future of autonomous robots. It's the present. A present that's still imperfect, that falls, makes mistakes, and needs fixing. But it walks. And when it needs to, it flies. And when it's convenient, it drives.
How long will it take to do it perfectly? Let's see.
