Spatial memory and learning: Towards a set of metrics to evaluate task performance in rats and robots

摘要

The study of spatial memory and learning in rats has inspired the development of multiple computational models that have lead to novel robotics architectures. Evaluation of computational models and resulting robotic architecture is usually carried out at the behavioral level by evaluating experimental tasks similar to those performed with rats. While multiple metrics are defined to evaluate behavioral performance in rats, metrics for robot task evaluation are very limited mostly to success/failure and time to complete task. In this paper we present a set of metrics taken from rat spatial memory and learning evaluation to further analyze performance in robots. The proposed set of metrics, learning latency and ability to navigate minimal distance to goal, should offer the robotics community additional tools to assess performance and validity of models in biologically-inspired robotic architectures at the task performance level. We also provide a comparative evaluation using these metrics between similar spatial tasks performed by rat and robot in comparable environments.