Robot Models of Human Behavior

Robots need models of human behavior for both inferring human goals and preferences, and predicting what people will do. Typical models, like the Boltzmann noisily-rational decision model, assume that people approximately optimize a reward function and choose trajectories in proportion to their exponentiated reward. In this work, we observe that these models originate from discrete decision making in econometrics, and propose, instead, a probabilistic model of human behavior for continuous trajectory spaces. Rather than each candidate trajectory affecting the human's decision independently, similar trajectories now affect the decision together.