Defence robotics

Research Lead/s :  
Prof. Robert Fitch

The future of robotics is in teams of robots that can work together to achieve a shared goal. For example, in a disaster relief scenario searching for trapped survivors, a team of robots can search an area much faster and more reliably than a single robot can. If one member of the team gets stuck, or suffers a communications failure, the rest of the team can continue searching while the disabled robot gets back online.  

To realize this capability, we need the robots to be capable of cooperating and managing tasks between themselves. One way to do that is to have a “leader” robot that tells each team member what to do. This can work well in safe environments with reliable communications, but in dangerous situations with unreliable communications, this is a risky approach. If the leader robot loses communications or is disabled, the whole team stops.  

In contrast, our work is on researching decentralised, or “leaderless” cooperation algorithms for teams of robots, in which team members decide among themselves what to do next. The removal of the “leader” role means that the failure of any member of the team will not stop the rest of the team from continuing to perform the task. Remarkably, our decentralised algorithms still maintain the same optimality guarantees as the case with a single leader, meaning that the team’s performance of the task is maintained. This work has applications in not only in humanitarian aid and disaster relief, but other search-and-rescue situations and also military scenarios. This work is a collaboration with several partners, including the DSTG. Our work in this area has been recognised with a Best Poster Award at the ICRA 2021 workshop “Robotic Swarms for the Real World”.  

Research Team : Felix Kong, Chanyeol Yoo, Brian Lee, Cadmus To, Fred Sukkar & Jennifer Wakulicz 

Research Strength : Sensing, perception and estimation, Control, planning and coordination, Platforms

Research Lead/s :  
Prof. Robert Fitch

Underwater gliders are a kind of marine vehicle that can remain at sea for weeks or months, which makes them suited for long-duration defence applications in addition to high-value missions in marine science and engineering.

One of our long-term goals is to enable these gliders to be autonomous by developing path planning algorithms for them, allowing fast and accurate navigation through the ocean with little or no supervision. 

Our experience in operating gliders in ocean trials has given us a glimpse of the difficult challenges that stand in the way of achieving our goal. In order to achieve long mission times, gliders avoid the use of active propulsion (e.g., propellers) and thus have extremely limited thrust. This means that ocean currents have a dominant effect on the glider's motion. If we understand how to accurately estimate and exploit ocean current patterns in designing planning algorithms, they can be used to the glider’s advantage. If poorly estimated or neglected, they can result in suboptimal plans, or worse, carry the robot hopelessly off course. 

The focus of our work is driven by the need for high-fidelity underwater current estimation that supports stochastic planning. We have developed algorithms to calculate fast and robust trajectories using sampling-based planning combined with ideas from areas such as formal methods and fluid dynamics. We have also begun to develop algorithms for real-time estimation of ocean currents that use ideas from control theory such as dynamic mode decomposition to augment noisy forecasts with local observations. This work is done in collaboration with several partners including the DSTG and the Australian Bureau of Meteorology. 

Research Team : Felix Kong, Chanyeol Yoo, James Lee, Cadmus To, Brian Lee, Giovanni D'Urso & Jennifer Wakulicz 

Research Strength : Sensing, perception and estimation, Control, planning and coordination, Platforms