Abstract
Service robots are integrating more and more into our daily lives to help uswith various tasks. In such environments, robots frequently face new objectswhile working in the environment and need to learn them in an open-endedfashion. Furthermore, such robots must be able to recognize a wide range ofobject categories. In this paper, we present a lifelong ensemble learningapproach based on multiple representations to address the few-shot objectrecognition problem. In particular, we form ensemble methods based on deeprepresentations and handcrafted 3D shape descriptors. To facilitate lifelonglearning, each approach is equipped with a memory unit for storing andretrieving object information instantly. The proposed model is suitable foropen-ended learning scenarios where the number of 3D object categories is notfixed and can grow over time. We have performed extensive sets of experimentsto assess the performance of the proposed approach in offline, and open-endedscenarios. For the evaluation purpose, in addition to real object datasets, wegenerate a large synthetic household objects dataset consisting of 27000 viewsof 90 objects. Experimental results demonstrate the effectiveness of theproposed method on 3D object recognition tasks, as well as its superiorperformance over the state-of-the-art approaches. Additionally, we demonstratedthe effectiveness of our approach in both simulated and real-robot settings,where the robot rapidly learned new categories from limited examples.