Ricerc@Sapienza

In this work, we have investigated the concept of “restraining bolt”, inspired by Science Fiction. We have two distinct sets of features extracted from the world, one by the agent and one by the authority imposing some restraining specifications on the behaviour of the agent (the “restraining bolt”). The two sets of features and, hence the model of the world attainable from them, are apparently unrelated since of interest to independent parties. However, they both account for (aspects of) the same world.

A common problem in Reinforcement Learning (RL) is that the reward function is hard to express. This can be overcome by resorting to Inverse Reinforcement Learning (IRL), which consists in first obtaining a reward function from a set of execution traces generated by the expert agent, and then making the learning agent learn the expert's behavior --this is known as Transfer Learning (TL). Typical IRL solutions rely on a numerical representation of the reward function, which raises problems related to the adopted optimization procedures.