UAV Navigation using semantic cues

Project dates: 25/02/2019 - Ongoing

Why does it matter?

Semantic cues are a natural way of navigating for humans, however, their incorporation for autonomous vehicles is limited. UAVs especially are hindered by size, weight and computational constraints, making the direct translation of contemporary research difficult.

Overview

This project aims to incorporate semantic cues into the navigation pipeline of UAVs. One of the projects key aspects is the influence of the unique positions of UAVs on the semantic information extracted from the vision system. Another key aspect is the transferral of algorithms to the constrained computing environment provided by UAVs.

Real World Impact

UAVs are becoming more ubiquitous and aid humans in solving a variety of problems, such as search and rescue, monitoring, inspection and other tasks. The majority of these tasks are concerned with reaching a position in space relative to an object of interest, with traditional algorithms focusing on topological or metric representations of the environment. Semantic navigation is the state of the art navigation performance exhibited by humans and navigation algorithms struggle achieving comparable performance.


Publications