"Common sense" used to be a term used often in AI research, although seldom implemented. Collecting large numbers of factoids has been done (see Cyc and ConceptNet). That didn't lead to common sense, although it helps some quesion-answering systems.
A useful working definition is that common sense is the predictive capability needed to get through the next seconds to a minute of life. This is what you need to not fall down, get hit by moving objects, and avoid predators. If you can't manage that, life will be short. It's a basic function of animal behavior.
Not enough has been done in that area. It's essential for robots. Google's self-driving cars have something like that, trying to predict what other cars are going to do in the immediate future.
One way to work on this problem is to have a system that looks at movie clips, and is asked to predict what happens next. It can, of course, be trained by looking at what does happen next. Image analysis is now far enough along to attempt this. Anyone doing that in recent years?