Probabilistic Semantic Data Association for Collaborative Human-Robot Sensing

Humans cannot always be treated as oracles for collaborative sensing. Robots thus need to maintain beliefs over unknown world states when receiving semantic data from humans, well account possible discrepancies between human-provided and these beliefs. To this end, paper introduces the problem of association (SDA) in relation conventional problems sensor fusion. It then develops a novel probabilistic (PSDA) algorithm rigorously address SDA general settings, unlike previous work on fusion which developed heuristic techniques specific settings. PSDA is further incorporated into recursive hybrid Bayesian scheme uses Gaussian mixture priors object softmax functions human likelihoods. Simulations multi-object search task show that enables robust state estimation under wide range conditions where can erroneous or contain significant reference ambiguities.