The Road Less Traveled: Decision-Making for Autonomous Mobility Using Remotely Sensed Terrain Parameters in Off-Road Environments

Case study was performed in 2020.

Presented by (TA1): Paramsothy Jayakumar (U.S. Army GVSC), Thomas Oommen, Jeremy Bos (MTU), Vijitashwa Pandey (Oakland U.), Jordan Ewing (student MTU) and Christopher Slon (post-doc OU)

Autonomous ground vehicles are increasingly being deployed in both military and commercial applications. The optimal operation of these vehicles is a challenging endeavor and requires a confluence of many disparate factors. One must accurately model the environmental conditions under limited information, such as terrain characteristics, reconcile them with vehicle capabilities, while simultaneously addressing mission requirements. Mission requirements typically involve attributes such as time to completion, vehicle speed, energy consumption, among others – and tend to be the operator’s main focus. For military operations, obtaining terrain properties can be difficult or impossible in war-zones and inaccessible territories using traditional ground data collection methods. Even when accessible, the data collected by the traditional approach can be sparse, leading to significant uncertainty which complicates path planning and mission related decision making.

This case study presentation provides an integrated decision theory based approach to accomplish these tasks. Terrain characteristics are acquired remotely using hyperspectral and thermal sensors mounted on an unmanned aerial vehicle. This enables predictions of soil properties such as soil type and strength over large terrain. The terrain information is then cascaded to a terrain- and platform-aware path planner that provides a traversable path. This path-planner has been characterized in multiple sensor configurations allowing the calculation of mission attributes. A multi-attribute utility function is calculated next so that optimal decisions can be made. This utility-maximization step ensures that operator preferences directly influence every vehicle-level decision made. It further allows for analyses such as, finding optimal parameters for the cost function used in path planning, and also the value of information acquired from the remote terrain sensing. The case study is demonstrated using data acquired on a Husky robot at the Keweenaw Research Center of the Michigan Technological University. Future research directions in all three areas of remote terrain sensing, path planning and utility based mission planning are also identified.