Continual Learning of Visual Concepts for Robots through Limited Supervision

TL;DR

This paper presents methods enabling robots to continually learn new visual concepts in dynamic environments with limited supervision, effectively remembering past knowledge and applying it to acquire new objects and scenes.

Contribution

It introduces models that facilitate continual visual learning in robots, addressing data scarcity and environment variability, with demonstrated success on benchmark datasets and real-world scenarios.

Findings

Effective continual learning in robots demonstrated on benchmarks

Robots can learn new objects and scenes in unconstrained environments

Models retain previous knowledge while acquiring new concepts

Abstract

For many real-world robotics applications, robots need to continually adapt and learn new concepts. Further, robots need to learn through limited data because of scarcity of labeled data in the real-world environments. To this end, my research focuses on developing robots that continually learn in dynamic unseen environments/scenarios, learn from limited human supervision, remember previously learned knowledge and use that knowledge to learn new concepts. I develop machine learning models that not only produce State-of-the-results on benchmark datasets but also allow robots to learn new objects and scenes in unconstrained environments which lead to a variety of novel robotics applications.