It just goes to show that AI enabled machines, especially the little robot vacuum cleaners we have in our homes, are not just becoming more normal but are actually seeing a rise in sales... more than likely to do with the fact they work... very, very well.

We've had a Roomba for a few years now but it was only when doing the Reinforcement Learning and Reward Engineering module in my recent AI strategy course at The Berkeley Institute, did I actually understand how clever this little bundle of joy is.

For an application like this, the following is all taking place:

  • States (all possible configurations that the robot and environment can be in). 
  • Actions (choices available to the robot at a given time).
  • Transition modelling (how the robot and environment change over time when actions are taken), and
  • Reward functions (the way the robot gets scored on its performance by assigning numerical values to each possible outcome – higher is better).

Clever little thing!

Next time you look at your little robot vacuum cleaner (if you have one), just imagine all of this taking place, with the robot learning continuously!

Who'd have thunk it... eh?