Visual inertial odometry is an uprising approach for robotics localisation as an alternative to the more mature and conventional LiDAR-based localisation. Inevitably, a shortcoming of vision perception is that it has a shorter perception than LiDAR’s, making it often less reliable in vast areas. However, visual perception still appeals to a large audience of technical users because of image sequences contain huge amount of information which have many other potential uses.

This project is established for a specialised field that focuses on vision perception solutions for mobile robots to achieve autonomous navigation capabilities in GPS-denied environments.

The target and emphasis are to grow a suite of lightweight vision sensors and further develop an advanced vision perception software stack.

Hence, in this project, the team’s primary aim is to fuse object instance detection results with the vision-based odometry to achieve optimal localisation with little or no drift in commercial settings.