This robot prefers Jenga over Go (Photo from MIT)
Jenga may be one of the most popular and simple games to play; you keep the tower going until it tips over. But it actually takes a good amount of dexterity, patience, and physical manipulation to master it. These lessons one MIT robot is learning. In their newly published study, scientists from MIT write about teaching a real-world robot physics and a practical sense of touch by having it play Jenga for hours on end.
The robot, developed by MIT, has a soft-pronged gripper, a force-sensing wrist cuff, and an external camera. The bot uses all of these to see and feel the tower along with each and every block. While the robot plays, a computer records the visual and tactile feedback from its camera and cuff and compares the measurements to moves that robot previously made. It also thinks about the outcomes of the moves, whether a block placed a certain way and pushed with a certain amount of force was successfully extracted or not. All this works together to allow the robot to learn in real time whether to keep pushing the block or move on to a different one.
But why Jenga? Because Jenga requires “mastery of physical skills such as probing, pushing, pulling, placing, and aligning pieces.” It’s an interactive experience where in order to move forward, you have to physically touch the tower to figure out how and when to move the blocks. This differs from purely cognitive games like chess or Go both of which rely on visual cues.
Instead of carrying out thousands of attempts, the bot was trained on roughly 300, with efforts of similar measurements and outcomes grouped by clusters that represent certain block behaviors. The cluster is key since it increases the efficiency with which the robot can learn to play the game and is inspired by the way humans cluster similar behavior. "The robot builds clusters and then learns models for each of these clusters, instead of learning a model that captures absolutely everything that could happen," said the paper's lead author, Nima Fazeli.
For now, the team isn’t interested in creating the next robotic Jenga champion. Instead, they’re interested in how this new learning system can be applied in other applications. The team thinks it’ll be especially useful in tasks that need careful physical interaction, such as separating recyclable objects from landfill trash and assembling consumer products.
“There are many tasks that we do with our hands where the feeling of doing it ‘the right way’ comes in the language of forces and tactile cues,” Rodriguez says. “For tasks like these, a similar approach to ours could figure it out.”
Have a story tip? Message me at: cabe(at)element14(dot)com