Summary
The project investigates how to make intelligent robots more reliable, understandable, and safe when performing complex tasks over long periods of time. Modern learning-based systems can achieve impressive performance, but they often lack guarantees about correctness and safety, especially when tasks involve sequencing, repetition, and conditional behavior. The project addresses these challenges by combining robot learning with rigorous reasoning techniques. The project's novelties are a formulation that embeds temporal logic task constraints directly into the reinforcement learning objective
