The Darpa SubT Challenge

An Interview with one of the winners of the World’s Hardest Robotics Challenge‍

The DARPA-funded Subterranean “SubT” robotics challenge has set out to revolutionize the autonomous exploration operations in the underground domain. World-leading robotics teams have been competing to deploy the most performant systems and win the $2 million prize money. The winning team, CERBERUS, used legged and aerial robots to tackle the harsh underground environments.

Learn from  Marco Tranzatto, the technical lead for the Robotics Systems Lab of ETH Zürich about his team’s approach on how to win one of the world’s hardest and most prestigious robotics challenges.

Let’s start with the problem you solved. Can you summarize what the competition was about?

The DARPA Subterranean Challenge has been a 3 year-long robotics competition initiated by the DARPA Agency, the biggest research agency in the US. It aimed to challenge the best robotics teams in the world to seek novel approaches to map, navigate and explore underground locations. In particular, DARPA wanted to see how robots can explore such challenging environments, support first responders, and provide rescue teams with a better situational awareness of what is happening underground.

The competition itself was split into different tracks: Tunnel Circuit, Urban circuit, and Cave circuit. It was very interesting to see how robots can adapt to diverse and difficult environments.

What was your role in this challenge?‍

I work as a robotics engineer at the Robotics Systems Lab at ETH Zurich, led by Prof. Dr. Marco Hutter. Together with other robotics groups, including the Autonomous Systems Lab at ETH Zurich, the Norwegian University of Science and Technology, and Oxford University, we formed the Team CERBERUS. My role in the competition was the Technical Lead for the Robotics Systems Lab.

‍

‍How did you approach the problem?

Team CERBERUS wanted to formulate a general solution to the problem of robotics exploration. The idea was to combine legged and aerial robots to explore this environment and use the best out of both. For example, legged robots provide very robust locomotion and adapt to different types of terrains, whereas aerial robots can explore vertical shafts and work faster. 

The main idea was that even though these robots are very diverse, overall they should have the same core functionalities, which in my personal opinion was one of the main strengths of our team.

What do you think was the biggest difficulty for your team during the project?

There were many challenges. But one of the biggest ones was that the team was distributed all over the world, some team members were working in the US, others in Norway and Zurich. But, at the same time, this brought a lot of diversity to the team which resulted to be also a big asset for us.

How about Sevensense? Can you share how our technology helped you win the challenge?

All the legged robots we used, all Anymal robots by Anybotics, were equipped with Core Research, the visual-inertial sensor by Sevensense, which offers a combination of the much-needed features: multi-camera support, synchronized visual-inertial data, high-sensitivity image sensors, and robotics-specific exposure algorithms.

We then built a ruggedized case for it to make sure that it was robust and ready to be deployed in underground environments where it could resist water and dust. Overall, the Core Research was very important for our success in the SubT challenge because it not only gave us sensor data for the visual-inertial odometry pipeline but also allowed us to have a surround-view by placing the cameras in different positions around the robot. The goal of the SubT challenge was to look and locate objects of interest, such as manikins that were mimicking potential survivors. Having such a broad field of view was a crucial point in scoring that high in the final competition.

From a technical point of view, it was also very important to have a robust and reliable sensor that was ready to be deployed and easy to be integrated. So, overall I would say the Core Research was a key component for us in winning the competition.

Overall, the Core Research was very important for our success in the SubT challenge because it not only gave us sensor data for the visual-inertial odometry pipeline but also allowed us to have a surround-view by placing the cameras in different positions around the robot.

What do you think was the biggest strength of your team?

Team CERBERUS was composed of several research groups from different universities, which in the end turned out to be the biggest challenge and strength at the same time. Everybody brought their expertise to the team. So it was great to have such an heterogeneous team of people, where everybody was pushing and working hard in his/her own field to achieve a common goal. This was an essential part of our Team and played a big role in our success.

What are you most proud of (except actually winning of course :D)?‍

Again, working with such an amazing team and great people. It was great to see how much passion everyone puts into their work. It was a long project and sometimes very stressful, but it was great to see how much everyone has been pushing towards this common goal.

Thanks to Marco Tranzatto for this insightful and interesting conversation. Watch the video below to learn more about the challenge.

‍