Within the next few years, the answer could be a team of robots. And if the success of CSIRO’s Data61 team in final of the DARPA Subterranean Challenge (also dubbed the Robot Olympics) is anything to go by, a lot of the software and hardware will have been developed by CSIRO and its partners.
CSIRO Data61 was one of 11 teams worldwide selected to participate in the three-year challenge, one of eight to reach the final held in September, and came second, pipped by 46 seconds.
In the process, they eclipsed teams involving organisations well known for their robotic expertise, such as Carnegie-Mellon University and NASA’s Jet Propulsion Lab (JPL).
“This is an amazing result! It’s a victory for all of Australia and cements CSIRO’s place as a world leader in robotics,” says Dr Navinda Kottege, who heads CSIRO’s Robotics and Autonomous Systems Group and led the Challenge team.
The DARPA challenge
The Defense Advanced Research Projects Agency (DARPA) has long been known for its innovative approach to the development of new technology.
The $US82-million Subterranean Challenge, for instance, was a competition with prizes of $US2 million for the winner and $US1 million for the runner-up.
The initial stages comprised three circuit events held in a tunnel system (a coal mine near Pittsburgh), an urban underground environment (an uncommissioned nuclear power station in Washington State), and a natural cave system (cancelled due to COVID-19).
The final, which combined elements of all three environments, was held in the Louisville Mega Cavern, Kentucky.
Contestants scored points for determining the position of and classifying objects of 10 different objects that might be found in the aftermath of a natural disaster or industrial accident.
This included backpacks, mobile phones, fire extinguishers, pockets of gas (CO2) and survivors (human dummies).
The underground environment where the events took place was unseen by the teams, unmapped, and impenetrable to GPS and radio signals from the surface.
Only one person from each team was allowed to communicate with and control the robots.
Because the terrain was likely to be difficult and varied – muddy, slippery, uneven, containing stairs, steep drops, and large caverns – CSIRO Data61 decided to use a mixed group of two tracked and two legged ground robots together with two drones.
The drones, which proved useful for exploring areas with high ceilings and precipitous drops have a limited flight time, and had to be carried by a tracked robot until deployed.
CSIRO’s Data61 team’s ace-in-the-hole was its world-leading 3D simultaneous localisation and mapping (SLAM) technology known as Wildcat, which the CSIRO has been developing for 15 years.
The platform’s algorithm combines data from a spinning light detecting and ranging (LiDAR) sensor and a sophisticated inertial measurement unit (IMU) to allow robots to map the surrounding environment.
On the ground robots, Wildcat SLAM with its LiDAR and IMU was combined with cameras for the detection of objects into a package known as Catpack.
An equivalent package without the cameras, Hovermap, was onboard the Emesentdrones.
Winners in accurate reporting
The similarity of the technology on each robot allowed data to be integrated seamlessly, and helped the team to win the “Single Most Accurate Report” award at the Challenge.
But the team of robots also had to be equipped with a communication system that could swap information between members underground and relay it back to the human controller on the surface.
What CSIRO Data61 developed was like a breadcrumb trail of communication hubs that could be dropped off from a tracked robot along its path.
The robots also needed enough autonomy to perform tasks on their own, out of reach of communication, and then return to report on what they had done or detected.
And the tasks themselves had to be allocated automatically in an efficient way, divided up so that no two robots rushed to work on the same thing at the same time.
The task allocation system was developed by researchers at one of CSIRO Data61’s partners, the Georgia Institute of Technology.
“The whole point of this work,” says Navinda, “is to develop utterly reliable technology to take humans out of dull, dangerous and dirty industrial situations, such as inspections in constrained spaces, checking out valves or pipes or sewer systems for cracks, for instance.”
“The market and industry are perhaps not yet ready to adopt some of the robot autonomy, coordination and control technology we developed for the Challenge,” says Dr Farid Kendoul, the chief technical officer and co-founder of Emesent, one of CSIRO’s partners in the team, “but it’s a natural progression.” Emesent, a CSIRO spinout, specialises in collecting and analysing information from challenging environments and provided the drones used in the Challenge.
“We have clients in mining, power plants, the oil and gas industry, and construction,” says Farid.
“They are blown away by the technology, because it can give them access to where they couldn’t get access before, it automates their workflows and saves them money, and allows their staff to work safely in challenging environments.”
The proof of Emesent’s success is in its employment growth, from the two founders in 2018 to 125 people today.
There is still plenty of work to do, Navinda says.
On to the next navigation challenge
“One of the hardest problems for robots is to navigate through natural environments, like forests and grasslands.”
But that has particular application in assisting future agriculture as it switches from the existing monoculture approach to polyculture, farming that generates many different products simultaneously from the same plot of land.
The $US1 million ($AUD1.3million) prize money will be reinvested into CSIRO Data61’s research and development of Australian technology.