Active Search In the Wild

Having robots be able to conduct visual searching of objects in their environment is an important step towards building intelligent robotic assistants. Automatic object searching is a useful tool for social, service, and search and rescue robots; such robots would be required to reason about likely locations of an object specified by a user. However, much of the existing work in the area of automated visual search assumes the target object to be in the detection dataset category, even though the dataset has a limited category of annotations. To mitigate this problem, we propose a project which aims to search for a query object that is known seen in the dataset. We will first propose an object detection model to detect an unknown object that is not in the detection dataset, and calculate its similarity to a user-queried word. Next, we aim to apply this detection model to generalize an object searching problem, where the query object can be any word. Finally, we will conduct experiments using this system to determine system functionality on real robot through human studies.

Faculty Supervisor:

Matthew Pan

Student:

Partner:

Korea University

Discipline:

Engineering

Sector:

Artificial Intelligence; Technology

University:

Queen's University

Program: