AI-powered
podcast player
Listen to all your favourite podcasts with AI-powered features
Train Assimilation, but You Can't Expect Generalization
You should be very cautious about any expectations on generalization from what you train on. The data you train on is going to determine how well it's going to work on data you put into the network. But if you train in similation where have many, many versions of the similator, then there is a high chance it might also work in the real world. That at least in our case, well for essentially identifying where objects are on a table,. It worked well for inhand manipulation of a rubic cube.