An Iowa Gambling Task-based experiment applied to robots: A Study on Long-term Decision Making

Abstract

Designing a robot’s decision-making process is challenging because it is still not completely understood even in humans. However, it is a fundamental process in the search for autonomous agents. When making decisions, we consider the short and long-term consequences of our actions, but some impairments prevent some people from seeing in the long run. Using as an inspiration an experiment carried out with humans in which decision-making is evaluated under the uncertainty of premises and results, rewards, and punishments, we created an equivalent robotics experiment. To model our agent’s state, we use a set of drives. Our agent’s goal is to reduce the distance between its homeostasis state and its needs. We trained a simulated robot with reinforcement learning, showing that long-term assessment agents can survive longer while satisfying other needs.