Abstract

One of the biggest concerns about collaborative robots (cobots) is their ability to adapt during their interactions with humans, which typically exhibit an immense diversity. We present an autonomous framework as a robot's real-time decision-making mechanism to anticipate and adapt to a great deal of human characteristics and behaviors, including human errors, toward a fluent and personalized human-robot collaboration. Our novel anticipatory architecture handles human behaviors in two levels: short-term context-dependent human behaviors, e.g., lost attention; long-term changing human characteristics, e.g., expertise. For the short-term adaptation, we implement a novel decision-making mechanism called anticipatory partially observable Markov decision process (A-POMDP). Our contribution lies in the ability of our model to handle the following human conditions to more accurately estimate the human?s need for help: 1) when the human's intention is estimated to be irrelevant to the assigned task, e.g., motivation is lost; 2) when the human does not want the robot's assistance in the given context, e.g., because of the human's distrust. For the long-term adaptation, we present a novel policy selection mechanism called Anticipatory Bayesian Policy Selection (ABPS), which is built on existing intention-aware models. ABPS selects from the set of policies that are generated from different A-POMDPs, each of which handles a different human?s short-term changing behaviors. The selection is based on the estimation of the human's long-term workplace characteristics, such as level of expertise, stamina, attention, and collaborativeness. We design a simulated factory environment consisting of crafted novel human behavior models collaborating at a conveyor belt with our robot to train our robot models and to conduct rigorous tests with greater uncertainty. Then, we integrate our solutions as an autonomous robotic framework and deploy it on a small-scale conveyor belt and a robot arm. Our novel setup presents a cognitively challenging task within a distractive environment to feature diverse human behaviors. We conduct user studies and the results show that our applied framework with extended human adaptation, covering changing human behaviors and their long-term characteristics, increases the efficiency and naturalness of the collaboration with a higher perceived collaboration, positive teammate traits, and human trust when compared to existing anticipatory solutions that do not handle such behaviors.