This video shows a simulation of the Parallel Hierarchical Replanner, one of the main contributions of my Ph.D. thesis. In the video, an aerial vehicle flies ahead of the ground vehicle and shares the terrain information it perceives. The ground vehicle builds a cost map from the combined information and adapts both its local and global plans in real-time.

We also did this for real, with me sitting in a helicopter discovering exactly how sick one can become when you have to look at a computer screen instead of the horizon...

This video is actually of another colleague's work (Alex Green), however it shows off the autonomous all-terrain navigation capabilities better than any of the pictures from this paper.

My main contribution was writing a version of the A* algorithm that could assess different costs depending on the orientation of the chosen edge in the graph.

The first half of this video shows the difference in performance between running my planning system without and then with prior terrain data obtained by a helicopter flyover. The screen captures show my operator GUI which indicates the overall paths taken in each case - the second one (with helicopter data) being far shorter and smoother. The second half of the video shows other people's work on the perception side of things.
  • Allen, T. Real-Time Path Planning for Autonomous Vehicles. Honours thesis, University of Sydney, 2006.

This video shows the Argo vehicle running the first real-time version my obstacle avoidance and path planning system. Raw data from the laser scanners was turned into an occupancy grid, resulting in the small tufts of grass being treated as solid obstacles. Net result? One 600kg autonomous all-terrain vehicle that carefully avoids treading on the grass.