Segmenting Robot Video into Actionable Subtasks

It's always hard to predict the next big leap in technology but it seems that companies will soon have a surplus of GPUs and RAM as their AI experiments come to an end. This will create some interesting opportunities for machine learning and, especially, robotics.

I think a major push we'll see soon is in the autonomous robots space. Things like self-driving cars, factory automation, and, more interestingly, home automation robots. There are a few efforts like Figure and Memo by Sunday to make general purpose humanoid robots. In the next 5 years I suspect we'll see a ramp-up in these efforts.

We've been dreaming of these bots since, at least, Karel Čapek's Rossum's Universal Robots play from 1920. The idea that one could have a machine do the cleaning and cooking has captivated the imagination of every generation. I feel we're very close to making this a reality.

Of course, as with all technology, dark patterns will emerge. Humans will become attached to these machines and anthropomorphize them. Imagine the prospect of having your emotional and physical needs taken care of by a thing that looks like an attractive 20 something year old who never ages, never argues with you, and never says no. Many countries are already experiencing negative population growth and if people start having romantic relationships with their bots this could become an epidemic.

Isn't this problem solved in the speech recognition domain by CTC? Nobody annotates phonemes there.