Even though Google is selling Boston Dynamics, that doesn’t mean it’s getting out of the robot business. Just ask its room full of robotic arms. It’s about more than just making a good robot arm, though. It’s also about teaching a robotic arm tactile learning skills.
Essentially, Google outfitted the arms with basic optical sensors and a neural network, or artificial brain capable of learning. The robot was then tasked with finding specific objects in a pile, and learning how best to grasp them in order to move the object. This is easier said than done, given that it’s a skill even humans gradually acquire. The researchers said it’s essentially hand-eye coordination.
The goal is a robot that could automate sorting tasks. This could be a bot designed for warehouse work, or one that could sort through single stream recycling to pick out glass bottles and aluminum cans from other items in the bin. Regardless, granting machines spatial reasoning and the ability to learn and adapt to objects could be a big boon going forward, with bots able to adjust its grasp for as hard or as light as the job necessitates.
Source: IEEE Spectrum