Science
Voices Only They Hear: Scientists May Finally Know the Source
05 November 2025
The alarm sounds, and evacuation begins. A squad of walking robots stands before the building. They can't reach the roof to neutralize the threat—stairs and narrow corridors will slow them down. That's why an army of drones launches from their backs. Sound like a scene from science fiction? Scientists at the California Institute of Technology (Caltech) just brought us closer to that scenario. Meet X1—a fusion of two robots that can move together in almost any environment, creating the ultimate multifunctional robot.
The X1 System is the result of three years of work aimed at merging two independent machines into one. This involves a humanoid walking robot and a flying drone that can also move on wheels when needed. Each of these devices has limitations on its own and cannot cope well in every situation. However, together, they create a nearly ideal duo. How did they achieve this integration?
The project utilized the G1 humanoid robot from Unitree. While its movement capabilities are impressive, it remains a relatively slow and unstable machine. Stairs, water obstacles, or narrow passages present a challenge to its rapid movement. Therefore, the scientists came up with an idea: use the G1 as a mobile landing platform for the drone.
The G1 carries the M4 drone on its back. This device is particularly interesting because, when necessary, its rotors can rotate 90 degrees and transform into wheels. Consequently, the system achieves full integration of three types of movement: walking, flying, and driving. That is why the X1 system is a true multifunctional robot.
As Caltech describes on its website, work on the system spanned three years. The team had to modify the humanoid G1 not only to mimic human movements but also to navigate the terrain independently. Teaching it to walk with the additional load on its back was crucial, as the mounted drone shifted the robot’s center of gravity.
What truly sets X1 apart from other robots is its method of learning movement. Most robots of this type operate on the principle of imitation: they record human gestures and scale them to their dimensions. X1 is different. The Caltech team created mathematical models describing the physics of movement and then combined them with machine learning techniques. The result?
A robot that understands the principles of walking at a fundamental level and can adapt to unknown situations. It doesn’t require pre-recorded patterns for every possible scenario. Moreover, when the need arises, it changes its mode of movement and, instead of walking further, launches the drone placed on its back.
The X1 robot was commanded to navigate a route consisting of sections requiring different techniques to overcome them. During the demonstration, the humanoid robot walked through the university library building, exited outside, and bent at the waist so the M4 drone could take off. The drone flew a short distance, landed, and transformed into a wheeled vehicle to efficiently cover the ground distance. When it encountered a pond, it ascended into the air again and flew over the obstacle.
Although the actions shown during the test seem straightforward, teaching them to X1 was not simple. The system was primarily continuously calculating the validity of its movements rather than relying on previously analyzed recordings of human movement. Complicated mathematical operations backed every step the robot took.
The next step in the research involves deeper integration of the G1 and M4 so they can autonomously adapt to their surroundings and react to changing conditions. Both devices must form an integrated whole when it comes to movement planning. Only then will this multifunctional robot system operate effectively.
Currently, the main question remains what application—beyond military or public order roles—flying-walking robots might have in our daily lives.
Read this article in Polish: Humanoidalny robot nie tylko chodzi i jeździ, ale także lata
Science
04 November 2025
Zmień tryb na ciemny