In this proposal, I present a research plan for semi-autonomous learning of a robotic assistant in a human-robot collaboration scenario. Under this framework, autonomous robots can intelligently gather data and ask questions from a human partner to improve their supportive capabilities while not overly detracting from human productivity. This framework is presented in the object detection domain, where adequate training data is difficult to come by. The proposed framework autonomously gathers relevant image data that it later asks human partners to disambiguate. Using active learning, the robot will ask the most informative set of limited questions for improving its classification performance. The approach also uses contextual information to guide search for relevant objects in an efficient manner. Examples of potential applications include EVA robotic assistants and personal satellite assistants.