Performance Metrics for State-Based Imitation Learning

Mohamad Zalat, Babak Esfandiari

Abstract: We propose five new domain-independent metrics for evaluating and comparing performance at imitating a state-based expert. We use two agents in the RoboCup environment to compare the performance metrics: an agent based on a Multi-Layer Perceptron (MLP) and an agent based on a Long Short-Term Memory (LSTM) neural network.