Active Visual Navigation in an Unexplored Environment

In this project, the goal is to develop a new method (using computer vision and machine learning techniques) for robotic navigation in which goals can be specified at a much higher level of abstraction than has previously been possible. This will be achieved using deep learning to make informed predictions about a scene layout and navigating as an active observer in which the predictions inform actions. The outcome will be robotic agents capable of effective and efficient navigation and operation in previously unseen environments, and the ability to control such agents with more human-like instructions. Such capabilities are desirable, and in some cases essential, for autonomous robots in a variety of important application areas including automated warehousing and high-level control of autonomous vehicles.

The aims of our project are therefore summarised as follows

Aim1: Develop representations and deep learning methods that capture the knowledge of scene layouts to enable prediction of high-level scene structure useful for local navigation as well as global planning

Aim2: Link higher-level scene representations to planning, creating methods to actively explore an environment and dynamically update imagined and real scene representations.

Aim3: Integrate Aims 1 and 2 to create a new method for robotic navigation based on high-level visual entities in the environment, enabling navigation in previously un-mapped environments