Online Bayesian changepoint detection for articulated motion models

摘要: We introduce CHAMP, an algorithm for online Bayesian changepoint detection in settings where it is difficult or undesirable to integrate over the parameters of candidate models. CHAMP used combination with several articulation models detect changes articulated motion objects world, allowing a robot infer physically-grounded task information. focus on three model appropriate: intrinsic relationships that can change time, object-object contact results quasi-static motion, and assembly tasks each step relationships. experimentally demonstrate this system be various types information from demonstration data including causal manipulation models, human-robot grasp correspondences, skill verification tests.