Evaluating Machine Learning Performance for Safe, Intelligent Robots
Raymond Sheh · 2021
The rapid advancement of Machine Learning techniques has been the primary driver of improvements in the performance of Intelligent Robots. Performance evaluation is a vital part of specifying requirements and evaluating capabilities for such systems. However, the performance evaluation techniques commonly used for machine learning systems must be augmented for use with robotic systems that interact with the real world. This paper presents a survey and discussion of factors that must be considered, beyond traditional measures such as cross-validation accuracy, when developing evaluation criteria for machine learned intelligent robotic systems.