Validation of virtual humanoid intelligent agents in virtual reality systems

One of the great benefits VR systems offer is their ability to simulate a number of virtual humans when their presence is needed in the context of some learning or training experience. Being that the real humans may not be available to play different roles and support virtual sessions, the ability of a system to generate highly believable representations of autonomous virtual humans - virtual intelligent agents - is vital in achieving specific learning and training objectives. Eliminating the elements of the system that can cause a negative learning and training transfer is a paramount in those systems. We illustrate the results of two user studies focused on validation of non-deterministic domain-specific behaviors generated by our system (example: behaviors typical for a well coordinated group of paramedics or military unit). The results and observations confirmed that when it comes to VR systems with stringent requirements and high expectations for positive learning/training transfer, we still need humans to evaluate and validate synthesized human-like agent behaviors.