Human Navigational Intent Inference with Probabilistic and Optimal Approaches

Presented at ICRA 2022, this paper addresses the problem of inferring where a pedestrian intends to go from partial trajectory observations — a core capability for robots navigating in shared human environments.

Problem

A robot sharing space with humans needs to predict not just where people are, but where they're going and why. Simple velocity extrapolation fails in complex environments where people adjust routes around obstacles, other people, and social norms. What's needed is inference over intent — the goal the person is trying to reach — not just trajectory continuation.

Framework

We formulate navigational intent inference as Bayesian goal inference: given a partial trajectory and a map, maintain a distribution over possible goal locations that explains the observed motion.

Likelihood model: We use an optimal planning model — a person heading to goal G would take a path close to optimal. The likelihood of an observed trajectory under goal G is proportional to how well it approximates the optimal path to G.

Prior over goals: Semantic priors from the map (exits, commonly-used destinations) combined with uniform uncertainty over novel environments.

Update rule: As trajectory observations accumulate, the posterior over goals sharpens. The model naturally handles mid-trajectory goal changes by allowing the posterior to shift.

Evaluation

We evaluated on datasets of pedestrian trajectories in indoor environments (building corridors, labs), measuring goal prediction accuracy at varying observation lengths. The probabilistic approach outperforms velocity-based and constant-velocity-model baselines, particularly at short observation lengths where intent is most ambiguous.

Relevance to Uncertainty in Robotics

This work connects to broader themes in my research: the importance of representing and communicating uncertainty rather than producing a single confident prediction. A robot that maintains a distribution over human goals can make better navigation decisions than one that commits to a single prediction and acts accordingly.