In recent years, virtual space simulations have been widely used to train responses to earthquakes, accidents, and vocational tasks. However, most conventional simulations only train one person at a time and do not teach the sequence of actions required in situations influenced by others. Non-immersive simulations, from an objective perspective, are considered effective for learning the order and priority of actions. Still, they are often difficult for users to operate, preventing them from focusing on learning. Additionally, communication opportunities that are essential for collaborative learning are lacking. This study proposes three approaches to enhance cooperative learning in non-immersive simulations to address these issues: introducing cooperative gestures to increase communication, supporting avatar operation through tutorials, and incorporating failure patterns for effective learning. Experimental results confirmed the effectiveness of these approaches in teaching appropriate actions and evacuation sequences during earthquakes, demonstrating their potential for enhancing collaborative learning.