Question: What is the hardest thing you build

Conor McGinn answered on 28 Feb 2016:

Hi,
That’s a really great question! Its a really tough thing to answer. Maybe a better answer would be to explain what are the hardest things about making ‘intelligent’ robots.
Ok so making a robot that someone can used in there home requires you to overcome problems on three levels; engineering, cognition and interaction.
The engineering problem relates to how the robot is constructed mechanically. We don’t build robots out of tissue and bone (yet) so we have to specifically design every aspect of it. Think about how complicated a person is and imagine how long it would take a person to build something that complicated! It needs to be roughly the same shape as a human (or it wont be able to do things like pass through doors or reach shelves), have arms/hands to grasp things, be safe in case it bumps into you, be power efficient so you dont have to constantly change the batteries etc. Solving these problems are incredibly difficult.
The second level of problems is in cognition – making the robot able to sense the world around it and make the correct decisions. The robot needs to have sensors in all the correct places. The sensors need to measure both the environment around the robot but also the they must measure the position of the robots joints. For example think of a robot trying to climb stairs – what sensors would it need to know where the stairs are and the position of its limbs relative to the stairs? Again we dont really have biology to help us much here so all the sensors we use are designed by people and are often times very limited in terms of what they can do. Also the robot doesn’t have a brain the same way as a person does. The robot’s brain is typically a computer that is programmed to behave a little like a brain. In general, regular computers make very poor substitutes for brains.
Finally the interaction problem… If we have a robot that can do what we want it to, and also behave the way we want it to, then we still need to figure out how to make it interact with people. Social interaction is a very complex problem in artificial intelligence. People are able to pick up meaning from very small behaviours – this is so so hard to make a robot do. For example, if someone slams a door a person can easily tell if it was intentional and if so, they can probably guess what the person might be feeling. People use sarcasm in conversations – they might say one thing and mean something else. Robots are so far away from reaching anything like this level of sophistication.
In conclusion, the biggest problem is finding a coherent way of putting a robot together that addresses each of these three problem areas. This is especially hard because in many cases we still don’t have answers for really important questions. And to make things harder, often when you introduce a feature on one problem level, it introduces problems on another level. For example – say you wanted to add an extra 3 sensors to make the cognition part better. That may affect the robots balance because these parts have mass (engineering level). Or it may make the robot look creepy and people wont want to interact with it (interaction level).
I hope this answers your question – finding a way to put all of this together is the hardest thing. Its an incredibly hard, but fascinating problem. And overall I think we are making great progress.