Development of a highly functional robot hand drive system that can 'solve Rubik's cube' is in progress
There are a lot of people who can solve Rubik's cube by rotating colored cubes and aligning the color of the face, but it is difficult for robots to solve Rubik's cube with machine hands and fingers. A large IT company, Tencent and a research team at Hong Kong University of China , announced that they have developed a system to solve Rubik's cube with a robot using AI.
Learning to Solve a Rubik's Cube with a Dexterous Hand
(PDF file) https://arxiv.org/pdf/1907.11388.pdf
AI researchers test a robot's dexterity by handling it a Rubik's cube
The fastest time for humans to solve Rubik's cube has been reduced gradually, and the world record at the time of writing is a whopping 3.47 seconds . This alone seems to be an extraordinary speed from the viewpoint of the general public, but in 2018, a machine that aligned Rubik's cube at such a speed that the eye of 0.38 seconds can not catch up has appeared.
The world's fastest 0.38 seconds machine solves the Rubik's cube, and a movie showing how to drastically change the world record is now available in GIGAZINE
However, this machine only uses special machines and motors, not machines with five fingers and joints like human beings. It is possible for humans to freely rotate and align Rubik's cubes freely using five fingers well, but it is difficult for robots to solve Rubik's cubes well using a structure with five fingers.
The researchers commented, 'It is important to operate your hands dexterously in order to perform various daily tasks where robots are as dexterous as humans and there are many contacts with objects.' We emphasize that if robots are able to perform activities using their hands in the same way as humans, they will be a major advance for robot development.
However, despite advances in the field of robots, many tasks remain in performing tasks using multiple fingers dexterously. According to researchers, a robot that can handle hands like human beings freely, such as the variety of patterns that five fingers touch with the target object, high dimensional work space, and fragile mechanical structure, etc. It seems to be a challenge on the top.
Therefore, the research team developed a model to find the optimal solution for 2x2x2 cubes ( pocket cubes ) with a smaller number of cubes than the usual 3x3x3. Furthermore, through robot hand simulation on virtual space, we developed a system to solve Rubik's cube with a robot hand.
Mujoco, a physics computing engine designed for research and development in the robotics and biomechanics fields, was used in this study. Among them, an experiment was conducted using a robot hand manipulator close to human hands called Shadow Hand . Shadow Hand is equipped with five fingers like human beings, and it is expressed that fingers with different length like thumb and little finger, moving joints and immovable joints.
The research team rotated the virtual Rubik's cube with the hand of Shadow Hand and solved it by combining the models leading to the solution of Rubik's cube. The driving model using AI seems to have worked very well, and the research team succeeded in rotating Rubik's cube with an accuracy of 95% or more, and achieved 90% or more of Rubik's cube completion in 1000 trials. It explains. The research team believes that the performance of the system can be improved in the future, and argued that the system could be deployed on real hardware as well.
by on xay