Effects of Trajectory Resolution on Human Intent Inference

Real-world human intent inference tasks are often performed online and demand high computational tractability. Human trajectories, on which intent inference algorithms rely, are thus approximated via discretized waypoints at a resolution of the designer's choice. The key contribution of this work is to investigate the effects of this trajectory resolution on the efficacy of intent inference algorithms. First, a Bayesian goal inference algorithm is formalized from an action-based perspective, and a novel method for discerning non-intentional motion is presented. Based on these frameworks, it is hypothesized that higher resolution trajectory representations enable better results for both goal and intentionality inference in near-optimal and noisy regimes. Finally, an experiment using human-generated trajectories was conducted to justify these claims.