MIT engineers have come up with a general code tool to optimise robot learning. They're calling it an «automated recipe for success,» one that can be applied to «virtually any autonomous robotic system» to accelerate the development of walking robots, self-driving cars, and other important robotics projects.
The standard process for robotics engineers is a monotonous one; there's a great deal of trial and error involved in perfecting robot movement(opens in new tab), as we've seen in the past. It's expected when engineers go into a robotics project that the AI will need to repeat the same movements over and over before it becomes even vaguely adept at completing a given task.
That process is made even more complex when it comes to deformable objects, but as we previously reported, MIT engineers are paving the way to highly complex AI with pizza dough-rolling robots(opens in new tab).
MIT News gave us the down-low on the project(opens in new tab) in which graduate student Charles Dawson, along with assistant professor in MIT’s Department of Aeronautics and Astronautics, ChuChu Fan, came up with the code in a bid to make the learning process less arduous for robots and their engineers alike.
In order to do this, they took the current way thinking and turned it upside down. As Dawson explains, «Instead of saying, 'Given a design, what's the performance?' we wanted to invert this to say, 'Given the performance we want to see, what is the design that gets us there?'»
From there they came up with the code using 'differentiable programming' techniques, which the study's abstract(opens in new tab) notes «can be used to automatically identify how and where to tweak a system to improve a robot's performance.» So rather than
Read more on pcgamer.com