Building interactive agents in video game worlds

Evaulaiting MultiModal Interactive Agents in Research: The Standardized Test Suite (STS) is a novel approach to evaluation of multi-modal interactive agents created by researchers Jaume Bosch et al. The STS uses behavioral scenarios from human interaction data and self-supervised learning to evaluate the agents’ performance against multiple objective criteria, such as accuracy, efficiency, and effectiveness. The paper examines the efficacy of the STS in evaluating various multi-modal interactive agents, including virtual robotics, chatbots, and personal assistants.