Special Issue on Deep Learning for Robotic Vision

Excerpt

In this special issue we consider the topic of Robotic Vision in the context of deep learning techniques. Vision for robotics, while closely related to classical computer vision, and often using it for inspiration, focuses on specific vision tasks which are required for robot navigation and interaction. It has distinct characteristics specific to the robotics domain, for example, working with multiple or diverse sensors, or incorporating sensing from different viewpoints. Robot vision has the opportunity to use self-supervision, e.g., the robot’s own actions or movements can be used as vision-based supervision. It also has its specific challenges related to robotics, for example, being able to perceive visual input in challenging non-structured environments, or in adverse weather conditions, being able transfer methods from simulated to real environments (i.e., sim2real transfer), and others. Robotic Vision is an indispensable part of robotics as many robotics-specific applications, such as learning to navigate, localize, or act in the environment, heavily rely on vision and thus are dependent on advances in automatic visual understanding. …