The DARPA challenge
The CSIRO robots – two drones and five tracked vehicles – competed in an international event that simulated potentially unsafe scenarios faced in mining, search and rescue, defence and even space exploration.
The competition challenges autonomous robots to find objects, such as mannequins, backpacks, fire extinguishers and mobile phones, and accurately report their locations – without GPS.
CSIRO is the only non-US lead organisation selected to compete in the event, known as the DARPA Subterranean (SubT) Challenge.
DARPA, the Defense Advanced Research Projects Agency, is the main US government funded defence research agency.
The competition, in Elma, Washington State, took place in an underground industrial environment at an unfinished nuclear power plant, which was fitted with corridors and barricades, with stairs and shafts connecting different levels.
No human competitor was allowed inside.
Bots with bells and whistles
The CSIRO robots have intelligent payloads, including a 'perception pack' featuring four cameras and a spinning laser that are the robot's eyes.
The robots use software developed by CSIRO, called Wildcat SLAM (simultaneous localisation and mapping), to generate a map of structures around them.
Another software program uses these maps to plot a safe pathway for the robot. This stops the robot from (for example) bumping into obstacles, trying to scale walls or falling off cliffs.
Only one human supervisor is allowed to communicate with the robots, so each robot needs to be capable of navigating itself.
"The robot has to go in there and figure out the location of doorways, passages, obstacles and open space," said Navinda Kottege, Research Team Leader and Senior Research Scientist with CSIRO’s Robotics and Autonomous Systems Group at the Queensland Centre for Advanced Technologies in Brisbane.
Another part of the SubT Challenge is for teams to be able to add sensors to the robots – for example, to detect gas leaks.
Mining robotics
Robotics has significant implications for the mining industry.
Robots could enter a collapsed mine after an explosion to tell if it is structurally sound and safe for people, and develop accurate maps of the structure to enable continuous mining.
Robots could also be useful in mine rehabilitation, enabling estimates of how much fill is required to be brought in, saving money and time and keeping people safe.
The DARPA SubT Challenge comprises tunnel (held in August 2019), urban (February 2020) and cave (August 2020) circuits.
The final, for which DARPA will select six competing teams, will be held in August 2021.
Focusing on the final
The CSIRO team, which is funded by DARPA and CSIRO, came fourth of 10 competitors in a tight urban competition.
It received a special mention for reporting the location of a backpack within 22 cm of its actual location.
The CSIRO robots detected backpacks, mobile phones, survivors (thermal mannequins), heated vents as well as gas leaks (CO2) in the four scored runs of the event.
Since finishing fifth of 11 in the tunnel challenge last year, the CSIRO team has been further developing the robots' hardware, software and communications.
"We have upgraded the chassis, motor and gearbox of the larger tracked robots, to make them lighter and improve weight distribution for easier handling and stair climbing," Dr Kottege said.
"And we have introduced communication nodes that give us much more robust and reliable bandwidth at a longer range, and enable multiple robots to share each other's map frames."
He said tracked robots cannot access areas such as mezzanine floors, and drones cannot access narrow entries and passageways.
The type of robots CSIRO used in the event depended on the type of environment encountered (DARPA provides no advance information on this).
Autonomous, self-locating, mapping and communicating
Atlas (pictured above) is a fully autonomous robot consisting of a Superdroid LTE2 platform as the base and CSIRO's Perception and autonomy pack on top.
This robot is capable of localising itself in an unknown environment while creating a map of the surrounding area using CSIRO's Wildcat SLAM technology.
It can then communicate this information back to a remote base station using wireless mesh networking.
The drones, which are carried on top of the larger tracked robots going into the course, have a shorter run time (about 15 minutes) compared to more than one hour for the tracked robots. The total time allowed for the course is one hour.
Points are earned if the robots correctly identify the type of object and the reported location is within five metres of the actual location.
Each course has a maximum of 20 objects.
Australian robotics in a global spotlight
The 20-member CSIRO team combines the expertise of CSIRO's Data61, spinoff Emesent, US research partner, the Georgia Institute of Technology, and other Australian partner universities and robotic companies.
"We are hoping to put Australian robotics capability under a global spotlight in this challenge, and also to use the technology that we develop to apply to problems faced by Australian industries where they require exploration in dangerous environments," Dr Kottege said.
The technology behind the robots – the perception pack, comprising the SLAM software, cameras and sensors – is in demand from companies who want to use it on their own vehicles.
CSIRO is in negotiation with potential industrial partners including oil and gas companies, the mining equipment, technology and services sector and the defence industry.