BSc Thesis (15HP): Embodied Conversational Agent

mbodied Conversational Agents (ECAs) have evolved significantly with recent advances in natural language processing and computer graphics. While frameworks like SAIBA provided a foundation for ECAs, modern AI tools offer new possibilities for creating more natural and engaging conversational agents. This project aims to develop a next-generation ECA system that leverages state-of-the-art language models and real-time character animation.

The project will explore the integration of on-premise DeepSeek language models for conversation generation, combined with modern behavior planning and gesture generation systems. The result will be a complete ECA system capable of natural conversation, appropriate emotional responses, and synchronized verbal and non-verbal behavior.

This interdisciplinary project combines aspects of natural language processing, real-time graphics, character animation, and distributed systems architecture. The work will contribute to understanding how modern AI tools can enhance human-agent interaction while maintaining privacy and system control through on-premise deployment.

Specific Tasks

• Set up and configure an on-premise DeepSeek language model for conversational AI
• Design and implement a modern interpretation of the SAIBA framework using current technologies (whisper, piper)
• Develop a 3D character in Unity with advanced animation capabilities (use existing models and animation sequences)
• Create a behavior planning system for generating appropriate gestures and expressions
• Implement real-time communication between the language model, behavior planner, and Unity character
• Document the system architecture and contribute to potential academic publication

The project is good for those interested in gaming and simulations and eager to learn Unity. Strong programming skill in either Python or C++ or C# or Java is a must.

The project is suitable for individuals or groups.

Contact

Andrey Kiselev (andrey.kiselev@oru.se)