KRISTINA Project – Initial Avatar Design


The GTI group has been working in Project Kristina, a Horizon202 EC-funded project that involves real-time interaction of a relational agent with older adults. Recently, at a project meeting in Ulm, our design of an avatar that can be used for initial testings was presented. The design presented several challenges regarding both technical and aesthetic requirements. The character was intended to maintain a neutral aspect in terms of ethnicity, but it was important also to develop an agent who can be empathetic and yet have authority.

There has been an emphasis in the development of a head because the facial expression is such a fundamental part of the project. The head has been built with a low poly mesh so that it can be processed and react in real-time as required. For the facial expression, a facial rig is being built to allow the avatar to be animated through bones and complemented with blend shapes. The bones provide additional mobility while the blend shapes are necessary so that the avatar can assume the necessary positions to be expressive and respond to a systematic lipsync.

This avatar is designed for initial testings and can be renewed, modified and updated according to further more advanced stages of the project.

Before starting the character design, the GTI carried out an analysis of state of the art for automatic avatar creation tools. They explain different approaches for avatar creation and mention important concepts and parameters that should be taken into account such as: concept of relational agents, difficulties for the process, principles of relational agent design, dialog structure and variability in agent behavior.  They also point out the importance of non-verbal behaviour and the ability to build and maintain long-term emotional relationships with users. Also, a comparison between three main softwares that simplify processes in creation of avatars, animation, editing and playback was stablished. Motionbuilder, Poser and MakeHuman were the softwares that were analysed. This research has given a broader perspective of what the project means. It has been a starting point to find out which can be a good workflow and which tools could eventually provide simplicity and effectiveness during the process.

Right now, the avatar needs more flexibility and different parameters to change basic characteristics such as it’s gender, age and general appearance. Further designs need to be made in the future so that the variability within the avatars can be proven to work under different circumstances.