Inferring and Learning Subgoal Sequences

Computational models for understanding human behavior are an important component for enabling robots to appropriately interact with people in shared environments. Recent techniques assume that behavior is goal-directed and infer the goals and future behaviors of people (Baker et al., 2006) to plan complementary robotic control (Ziebart et al., 2009; Kuderer et al., 2012) or learn imitative control policies (Verma & Rao, 2006; Michini et al., 2013).