Engineers in Switzerland have developed a four-legged robot capable of solving a giant Rubik’s cube. Most impressively, it performs the entire task autonomously, using its own limbs to manipulate and move the object. The demonstration was conducted by the Robotic Systems Lab at ETH Zurich using the ANYmal-D—a machine originally designed to carry out inspections in complex industrial environments.

Solving a Rubik’s cube is not particularly challenging for a computer, as algorithms exist that can calculate a solution in a matter of seconds. The real difficulty arises when a machine must physically execute each move—especially when it lacks fingers to grasp and rotate the pieces with precision. To overcome this obstacle, the researchers combined computer vision, motion planning, and reinforcement learning—an artificial intelligence technique in which a system learns through trial and reward.

First, the robot walks around the cube, using its cameras to scan the faces and identify the colors; when necessary, it rotates the object to reveal hidden sides. Once the full configuration has been mapped, an algorithm calculates the exact sequence of moves required to solve the puzzle. The most impressive part, however, occurs during execution: to rotate one of the layers, the robot must steady the cube by pressing one of its legs against it, while using another limb to push the pieces in the correct direction.

At the same time, it must calculate and distribute its own weight across its remaining legs to avoid losing its balance—since any misstep could shift the cube’s position and ruin the entire sequence. For this reason, engineers practiced these movements in virtual simulations before transferring them to an actual robot. During physical-world testing, the robot’s cameras and sensors continuously monitor the object’s position, allowing for real-time adjustments.

In the experiments presented, the robot managed to solve the cube in three attempts, taking between 8 and 9 minutes each time, with approximately 30 actions per solution. While this may seem like a highly sophisticated way to solve a puzzle, the core objective of the research goes well beyond mere entertainment. Quadrupedal robots are already capable of walking through factories, climbing stairs, and navigating difficult terrain; however, they typically require additional tools when they encounter an object that needs to be manipulated or moved out of the way.

This new approach demonstrates that the robot’s own legs can function as limbs for manipulation. In the future, this capability could allow industrial inspection robots to open doors, clear obstacles, or operate specific mechanisms—all without the need to carry heavy robotic arms. And it all began with a seemingly simple challenge: getting a robot to solve a Rubik’s cube without having any hands.