Abstract
Shared autonomy consists of a collaborative effort between a human user and a robotic system having a shared goal, in which the human-controlled robot adapts its behaviour to provide assistive actions. For effective assistance, shared autonomy systems are required to infer human goals and/or cognitive states from their inputs. Partially observable Markov decision processes (POMDPs) offer a rich mathematical framework for formulating the above-mentioned planning problems under uncertainty. This paper presents a systematic review on the reported applications of POMDPs in the shared-autonomy context. Various aspects of the current scientific literature have been analysed to determine common shared-autonomy scenarios, implementation of human parameters in POMDPs and assessment methodologies. The effectiveness and limitations of POMDP approach for shared autonomy systems are also discussed.