Kernel Modeling for Synthetic Personality

The paper proposes a kernel approach to the modeling of synthetic personalities. Each kernel, corresponding to a personality trait, is a collection of probability distributions suitable to describe how sensor data maps into features associated with personality traits. The linear combination of human personality traits, sometimes referred in the literature as being able to generate the richness of human personality, has a counterpart in the linear combination of the kernel modeled traits proposed in the paper. People detection data from a social robot in a real environment is used to demonstrate the estimation of the kernels.