Inferring User Intent using Bayesian Theory of Mind in Shared Avatar-Agent Virtual Environments.

Area of Science:

• Artificial Intelligence
• Human-Computer Interaction
• Cognitive Science

Background:

• Virtual environments require sophisticated agent behaviors for realistic interaction.
• Inferring user intentions is crucial for natural human-agent collaboration.
• Existing methods struggle with human irrationality and dynamic intentions.

Purpose of the Study:

• To develop a real-time algorithm for inferring user avatar intentions in multi-agent virtual environments.
• To enhance virtual agent responsiveness and behavior generation based on inferred user intent.
• To improve the user experience in immersive virtual settings.

Main Methods:

• Applied Bayesian Theory of Mind to infer hidden intentions from proxemics and gaze cues.
• Accounted for human behavioral irrationality and dynamic intention changes.
• Integrated inferred intent to guide virtual agent locomotion and gaze behaviors.

Main Results:

• The algorithm reliably and efficiently infers user intent, even under noisy conditions.
• Demonstrated significant reduction in false inferences compared to prior methods.
• User evaluations showed a 68.18% increase in comfort with agents using the new algorithm.

Conclusions:

• The developed algorithm provides a robust method for real-time intention inference in virtual reality.
• Improved intention inference leads to more plausible agent responses and enhanced user comfort.
• This approach facilitates more natural and immersive interactions with multiple virtual agents.