IPAB Workshop-08/10/2020

Title: Elaborating on Learned Demonstrations with Temporal Logic

Description: Most current methods for learning from demonstrations assume that those demonstrations alone are sufficient to learn the underlying task. This is often untrue, especially if extra safety specifications exist which were not present in the original demonstrations. In this talk, i'll show how an expert can elaborate on their original demonstration with additional specification information using linear temporal logic (LTL), then convert that LTL specification into a differentiable loss for learning. Further, by leveraging adversarial training, ill also show how we can robustly satisfy the given specification on inputs outside the training data.