Teaching robots to adjust their grip the way humans do

Pick up that can. Now say you had to put it down on its side rather than its base, how would you go about doing it? Very easily and without thinking about it, no doubt. But simple manipulations like that, which allow us to interact fluidly with objects and environments, are actually incredibly hard for robots to pull off. Researchers at MIT are working to make it a little bit easier.

Nikhil Chavan-Dafle is a mechanical engineering grad student who has been working for some time on allowing robots (a robot arm, specifically) to use the environment to their advantage when moving objects around.

Being able to improvise in order to complete a task is a great skill for a robot to have. If a piece slips or isn’t aligned perfectly when it’s picked up, the bot doesn’t freeze up and say “nope, can’t do it.”

“We want robots to exploit the environment and use it to change the pose of the object in hand,” Chavan-Dafle told TechCrunch. “We basically developed a formulation that allows robots to estimate how the forces and motions and contacts are going to be involved, and use this underlying model, it can predict how the object is going to move in the grasp.”

The team hopes to present their work at a conference later this year.