Question: what kind of things do you program robots to do ?

Conor McGinn answered on 8 Mar 2016:

For me, I program robots to do three things (1) work in houses (2) interact with people (3) control their own states. I can give you a short background to each of these areas:
1) Getting a robot to work in a home is really hard. Some of the things we’ve tried to get our robots to do ; automatically generate a map of a house that it has never been in before, answer the door when the bell goes and look for an object in a house. We are starting to focus now on manipulation and how we can make robots pick up things from counter tops etc.
2) This is really hard problem on several levels.
Its very hard to make a robot fully autonomous – plus often when a robot is fully autonomous people get a little worried that it will do something it shouldnt. So I am currently designing robots to be ‘semi-autonomous’. This means that they can be autonomous at some things but require manual control for others. In this case interaction is really important – just think about how frustrating it is to use a mobile phone that is faulty and you’ll have some idea. So in this case I program the robot to be able to perform autonomously, and also so that the robot can convert instructions from a human into actual motion commands.
Also I program the robot to interact with people in a social way. This requires integrating cool things like facial recognition, speech recognition etc into the robots functionality. I also program the robot to follow social cues – so for example when you approach the robot it knows to say ‘hi’ or it knows not to interrupt you when you are speaking. All if these kinds of things are really only in their infancy right now – you can expect much progress in the next 10 years in this area.
3) For a robot to move, even if it is under remote control you need software that interacts with actuators/motors. On my more recent robots, we have around 10 microntroller boards located throughout the robot. Their main purpose is figure out what position the robot is in (typically done by reading sensors in the joints). These boards are all interconnected so this information is then transmitted to a larger computer where a model of the robot is updated. Depending on the state of this model, and the task we want the robot to perform, this computer can then generate and send motion commands to the actuators.