In short, his argument is that we don’t know what the “killer app” will be for home robotics, so trying to build a robot based on a particular use case is a fool’s errand. He then describes Misty’s approach of providing a platform for others – companies, researchers, and hobbyists – to build on in search of that application.
Here’s the thing, though. We know exactly what that killer app is: home chores. There is a billion-dollar market for a robot that can…
Go anywhere in the home a human can. That includes opening doors and going up and down stairs, and that it not be stopped by ground clutter like toys or laundry. It must be cheap enough to have one per floor.
Manipulate the environment around it. I’d define that as being able to move objects of at least five pounds across a human range of motion.
Complete a wide variety of tasks. Being able to dust and do the laundry and dishes are not enough to justify the robots’ expense.
All of that is really hard to accomplish.
The locomotion requirement probably means it will walk instead of roll. The lifting requirement implies a large footprint and mass. Being general purpose will require further advancements in AI, but those will hopefully help reduce hardware costs at the same time as end-to-end reinforcement learning (or something similar) relaxes the need for ultra-precise mechanical components.
What we should say is, “We don’t know what a killer app for home robotics is that we can build today.” That brings into focus the problems we should be – and lots of people are
– striving to solve.