The University of Sydney’s Australian Centre for Field Robotics (ACFR) says it anticipates two more years of research and development effort is needed to achieve technical maturity for UAV guidance systems based on simultaneous localisation and mapping (SLAM) technologies.
ACFR has been flying increasingly sophisticated SLAM-based navigation suites for several years, some in co-operation with BAE Systems Australia on Australian Department of Defence-funded projects. The US Air Force has been actively funding elements of the ACFR research effort since midway through 2006.
SLAM technology is based on an autonomous UAV using its sensor suite to build up a map of its environment by cross referencing feature locations, with the results used to provide a navigation reference that can augment or replace GPS data in combat environments. Targeting quality maps can be produced using multiple UAVs sharing and processing data across a network.
Dr Salah Sukkarieh, associate professor at the University of Sydney, says that while SLAM technology is well advanced with respect to indoor navigation by robotic systems, its use in wide area conditions faces multiple challenges.
“SLAM works, but just. If you start to come along and look at hundreds of features, hundreds of targets, it starts to become computationally expensive so there is a lot of research going on now as to how you actually remove that computational burden for large-scale maps.
“Linking SLAM into control also leaves a difficult problem because your uncertainty keeps growing when you don’t see features that are out there and when you come along and re-observe a feature you will get a collapse in uncertainty or an increase in information and that could cause reverberations in the control solution.”
He says the technology also continues to experience issues with communications and the relative level of autonomous control capability supported by the air vehicle. “You have got limited communications bandwidth and you are trying to send these huge correlation matrices between platforms. The final one is this risk-modelling aspect – can you actually lesson the control burden by having the UAV make intelligent decisions about whether it should go out there and explore or come back in, depending on what the mission requirements are?”
ACFR plans to demonstrate a mature capability “probably within two years…It is probably going to be two years before we can actively demonstrate two UAVs doing active control on SLAM simultaneously.”
SLAM is expected to provide high-level autonomous localisation and navigation capabilities for military users, he says. “It opens up various form of operational benefits, especially for co-operative unmanned air systems, and that is where a lot of the interest comes in from the Department of Defense in the US.”
Near-term research by ACFR includes work on natural features extraction from an environment. “This poses a bigger issue because what happens is that you cannot appoint a centroid here. As you move around the centroids start to change. The question is how do you start to actually represent blobs of information and pass that into the data fusion algorithm? What that algorithm is doing is basically a feature extraction exercise where it can actually pull out trees from the environment and use that as part of the map.”
Source: FlightGlobal.com
