Machine learning has improved by leaps and bounds in the last half decade. Much of that progress has yet to make it out of the lab. But that's typical for new technologies, and will happen eventually.
It's not a huge technological leap from "robot learns to play Dota" to "robot learns to stitch baseball gloves". Machine vision and reinforcement learning were always the technologies holding back interesting applications of robotics. And now they are basically solved problems.
It's not a huge technological leap from "robot learns to play Dota" to "robot learns to stitch baseball gloves". Machine vision and reinforcement learning were always the technologies holding back interesting applications of robotics. And now they are basically solved problems.