MIT researchers’ new AI robot can beat you in the game of Jenga

Watch: MIT’s robot learns how to play the game Jenga. Researchers believe the robot can be used for manufacturing assembly lines in the future.

This AI robot knows how to play Jenga (MIT)

Jenga is one of the most popular indoor games. The slow-moving game truly tests your patience and focus. Soon, humans will have to compete with robots in the game of Jenga.

Researchers at MIT have developed a robotic arm that knows when to keep moving or switch to a new block without letting the tower fall. The arm is essentially a customised version of ABB IRB 120 robotic arm. It uses a soft gripper and force-sensing wrist joint to move the blocks while real-time learning from an external camera.

According to researchers, robot learned the game with mere 300 attempts instead of tens of thousands of such attempts which is expected to train an AI.

Alberto Rodriguez, the Walter Henry Gale Career Development Assistant Professor in the Department of Mechanical Engineering at MIT, pointed out that the robot stands out from previous milestones where robots could participate in cognitive games such as chess or Go.

"Unlike in more purely cognitive tasks or games such as chess or Go, playing the game of Jenga also requires mastery of physical skills such as probing, pushing, pulling, placing, and aligning pieces. It requires interactive perception and manipulation, where you have to go and touch the tower to learn how and when to move blocks," Rodriguez explained.

"This is very difficult to simulate, so the robot has to learn in the real world, by interacting with the real Jenga tower. The key challenge is to learn from a relatively small number of experiments by exploiting common sense about objects and physics," he added.

Researchers believe the Jenga playing robot could be used in the assembly and manufacturing industries.

"In a cellphone assembly line, in almost every single step, the feeling of a snap-fit, or a threaded screw, is coming from force and touch rather than vision. Learning models for those actions is prime real-estate for this kind of technology," said Rodriguez.