Explaining Reinforcement Learning Policies through Counterfactual Trajectories.
Julius FrostOlivia WatkinsEric WeinerPieter AbbeelTrevor DarrellBryan A. PlummerKate SaenkoPublished in: CoRR (2022)
Keyphrases
- reinforcement learning
- optimal policy
- policy search
- markov decision process
- control policies
- reward function
- fitted q iteration
- policy gradient methods
- reinforcement learning agents
- partially observable markov decision processes
- markov decision processes
- function approximation
- dynamic programming
- reinforcement learning algorithms
- state space
- hierarchical reinforcement learning
- moving object trajectories
- moving objects
- control policy
- trajectory data
- macro actions
- policy iteration
- markov decision problems
- total reward
- robotic control
- multi agent
- continuous state
- learning algorithm
- approximate policy iteration
- supervised learning
- temporal difference learning
- machine learning
- action selection
- model free
- decision problems
- learning process
- reinforcement learning methods
- policy gradient
- revenue management
- decentralized control
- spatio temporal
- average reward
- partially observable
- finite state
- learning problems