Appearance-based navigation for mobile robots

This work is a collaboration between the Intelligent Robotics Group and the Vision, Graphics and Visualisation Group, at AU.

This work is on building representations of (mobile) robot environments. The representations are not geometrical representations but rather visual. In other words, what is represented is the appearance of the environment as perceived by the robot.

The ultimate goal is to have a mobile robot exploring a new environment, building a representation of it, and using the representation to navigate from place to place. One use of the representation would be for a human to look at it (which should be straightforward since it is made of "images") and to tell the robot to go to a particular place, e.g. to further the exploration or perform a particular task. Another use would be for the robot to recognise particular places, e.g. docking stations, in the representation and to be able to navigate toward them.

Exploration of the environment and building and using the representation are typically concurrent tasks, i.e. the building happens while the robot explores and the representation is used to further the exploration, e.g. to find out places not yet explored. Moreover, the representation should be updated when a place previously explored is again visited.

One of the possible domains of application is planetary exploration but it is certainly not the only one. In particular, the work can be applied to tasks using semi-autonomous robots navigation, i.e. tasks where humans only provide very specific and/or high level input.

Another area could be along the line "Robot, here is an exemple of something you have to assemble. Do it now." In this case, the environment of the robot is simply a set bits that will make up the final product.

So far, we have results on building topological maps of the environment [NL04a, NL04b], on visual homing [ML04], on using images to simulate a compass [Lab04, Lab06, Lab06b] and on short and long-range visual navigation [BL06, Lab07]. It is all based on measuring the distance between images in a sub-symbolic way. More work is currently in progress and new results will be made available one day or another. Watch this space!