Having already wiped the floor with some of the world’s best in Chess, Go and doing plenty of damage to the collective morale of the Dota 2 and Starcraft communities, machine learning-powered AI now has a new game to flog hapless humans at: Jenga.
The robot, which is traditionally designed for use on the assembly line, was designed by MIT assistant professor Alberto Rodriguez and members of the university’s MCube lab. Using a range of sensors, it sees the Jenga tower as a mix of colours, which help it identify what blocks are the most structurally vulnerable, which ones are safest to move, and everything in between.
It’s a giant cheat, in other words.
Still, seeing the robot systematically work through the blocks is neat. It’d be nice to see how the machine learning model responds when someone tries to deliberately fuck with the Jenga tower, and what that would look like on the sensor grid. Theoretically, you could make a series of moves that means the robot has no other choice to take risky blocks – but ultimately, because it’s machine learning, it’d still have the advantage in identifying the right blocks.
Now if only someone could train an AI model to replace the people who keep taking Reinhardt in every Overwatch comp game I get matched into, that’d be sweet. If you want to nerd out over more stuff from MIT, their official YouTube channel for the Mechanical Engineering department has videos on harvesting clean drinking water out of thin air, a robot solving a Rubik’s Cube in .38 seconds, and a combat boot that uses kinetic energy from walking to power a GPS transmitter over the mobile network.
Leave a Reply