Multimodal operator decision models

This paper develops the multimodal softmax (MMS) model, a probability distribution for multimodal discrete random variables with continuous conditioning random variables. MMS is motivated by the problem of learning multimodal probabilities for categorical human decisions in Bayes Net models of semi-autonomous systems. The MMS model is then derived vis-a-vis softmax and softmax mixture distribution models. MMS training is discussed in the context of maximum likelihood estimation. Finally, decision classification results using experimental data from Cornell´s RoboFlag human-robotic interaction testbed are presented.