Estimating Levels of Engagement for Social Human-Robot Interaction using Legendre Memory Units

摘要

In this study, we examine whether the data requirements associated with training a system to recognize multiple 'levels' of an internal state can be reduced by training systems on the 'extremes' in a way that allows them to estimate "intermediate" classes as falling in-between the trained extremes. Specifically, this study explores whether a novel recurrent neural network, the Legendre Delay Network, added as a pre-processing step to a Multi-Layer Perception, produces an output which can be used to separate an untrained intermediate class of task engagement from the trained extreme classes. The results showed that identifying untrained classes after training on the extremes is feasible, particularly when using the Legendre Delay Network.