Overview / Usage

Reinforcement Learning (RL) is a powerful method of solving complex planning tasks. However, its real-world application is limited because the agent has to explore to learn by trial and error. Exploration is sometimes expensive or dangerous as errors can cause fatal consequences (e.g. in Autonomous Driving). One viable way to address the problem of safety during exploration is to do the exploration in a simulated environment. The goal of this project is to find a theoretically grounded method of transfer of trained policies from a simulator to the real world with confidence guarantees.