Robot dog solves Rubik's Cube within 9 minutes in Switzerland - Watch
&imwidth=600&imheight=450&format=webp&quality=medium "Robot dog solves Rubik's Cube within 9 minutes in Switzerland - Watch")
ANYmal can solve a Rubik's Cube in under 9 minutes. Photograph: (X/ ETH Zurich)
Scientists in Switzerland have trained a robot dog to solve the Rubik's Cube. The specimen used measured 40 centimetres per edge and weighed 4.6 kg. The robot needed to hold on to the puzzle while twisting the sides and also ensure its own balance.
Engineers at ETH Zurich in Switzerland have taught a robot dog to solve a giant Rubik’s Cube. The researchers say that the four-legged robot uses three legs to solve the puzzle, while using the fourth one to keep the cube steady. It has been named "ANYmal" and can solve the 4.6-kilogram Rubik's Cube in eight to nine minutes.
ETH research team said in the paper that ANYmal was able to successfully solve the puzzle three times without human assistance in trials. It carried out 30 movements each time, including the rotations and its own position changes. The cube measures 40 centimetres per edge.
ANYmal keeps the cube in place with its front paw and then uses the other legs to rotate two layers by 90 degrees. After this, it changes its position and moves on to the next step. But to be able to solve the puzzle, the robot needs to first identify the colours, which it does through its cameras. It captures images of the visible sides and then tips over the cube to see the colour on the underside. It uses its recordings to reconstruct the puzzle after the Kociemba algorithm calculates a solution that requires the least number of steps.
Balance and precision
The robot dog needed precise leg control to be able to complete the puzzle. This was essential since it had to not only ensure enough force to rotate the sides, but also ensure that it did not accidentally move the other sides, while maintaining balance.
For this, researchers used Reinforcement Learning, where an AI was used to first train the movement sequences. It started to learn to control its movements after repeated attempts and rewards. Using the results, the researchers developed a controller to work with ANYmal's images and sensor data.
While four-legged robots have used their legs for tasks that did not require a lot of manoeuvring, solving a Rubik's Cube was different. The robot had to carry out a set of precise and coordinated movements. Researchers say this has demonstrated the robot's ability to carry out tasks where it needs to hold, move and operate objects.
