MIT Engineers Develop A Robot That Can Play Jenga

A new robot from MIT uses machine learning algorithms to play Jenga. The robot analyzes its next moves through tactile and visual stimulation.  ( MIT Mechanical Engineering | YouTube )

In the past, robots and artificial intelligence have been winning against humans, at least when it comes to some video and board games.

Case in point: In August 2018, AI beat a team of pro Dota players in an exhibition match. Two years ago, AlphaGo AI, created by Google's DeepMind team, defeated humans in Chinese board game Go.

Now engineers at the Massachusetts Institute of Technology (MIT) have developed a robot that uses machine learning to play a game of Jenga.

Robot Learns How To Play Jenga

While other robots became masters at cognitive games such as chess, MIT researches trained a robot that can do something different. Unlike board games, which depends on visual cues, the new robot from MIT was trained to learn physical skills by analyzing visual and tactile feedback data in real-time.

"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," said Alberto Rodriguez, assistant professor at MIT's Department of Mechanical Engineering.

The researchers used a robotic arm equipped with a soft-pronged gripper along with a camera and a set of sensors. The robot uses all these to feel and visually analyze the whole tower and its individual blocks. When it pushes against the blocks, a computer gets feedback from the camera and sensors attached to the robot and compares them from the data of previous attempts. It also calculates possible outcomes before making a final move.

Instead of learning through tens of thousands of attempts, the researchers taught the robot using just 300 tries. This allows the robot to cluster its attempts and then study each of them, helping it to decide if a piece is stuck or free to move, which is somewhat similar to how humans learn.

"The key challenge is to learn from a relatively small number of experiments by exploiting common sense about objects and physics," Rodriguez explained.

Uses Of The Robot Beyond Playing Jenga

According to researchers, aside from training the robot to play Jenga, they are much more interested in developing it for other tasks that can be useful in the real word. This robot's manipulative skills look promising and could be beneficial in future industrial machines that are more accurate.

Further details of MIT's Jenga-playing robotic arm are published in the journal Science Robotics.

Watch the video below to see the robot in action:

ⓒ 2018 All rights reserved. Do not reproduce without permission.
Real Time Analytics