A review of reverse reinforcement learning at Virginia Tech (VT)

【 title 】A survey of inverse reinforcement learning（ A review of inverse reinforcement learning ）

【 The author team 】Stephen Adams, Tyler Cody, Peter A. Beling

【 Date of publication 】2022.2.8

【 Thesis link 】https://link.springer.com/content/pdf/10.1007/s10462-021-10108-x.pdf

【 Recommended reasons 】 Learn from the demonstration , Or imitate learning , It is the process of learning to act in the environment from the examples provided by the teacher . Reverse reinforcement learning （IRL） It is a special form of learning from demonstration , It attempts to estimate the reward function of Markov decision process from the example provided by the teacher . Reward function is usually considered to be the most concise description of a task . In simple applications , The reward function may be known , It may also be easy to export from the properties of the system , And hard code it into the learning process . However , In complex applications , This may not be feasible , And it may be easier to learn the reward function by observing the teacher's behavior . This paper deals with IRL The relevant literature is reviewed . The survey outlines IRL And the difference between two similar methods —— Apprenticeship and inverse optimal control . Besides , This survey is organized according to the main methods IRL The literature , It describes IRL Application of algorithm , It also provides areas for future research .