After an eventful spring filled with exciting development for the Reeds project, we are back with big updates. First of all, thanks to all who signed up to be part of the Reeds community. The interest has been huge and we are happy to say that we now have members from 16 countries!
Talking about the Reeds community: We are happy to announce that we will give our first public demonstration of the Reeds data at the Breaking the Surface conference, the 14th International Interdisciplinary Field Workshop of Maritime Robotics and Applications, in Croatia, September 25th to October 2nd (http://bts.fer.hr). Associate Professor Ola Benderius from Chalmers University of Technology and Ted Sjöblom from Research Institute of Sweden (RISE) will demonstrate the Reeds dataset in a lecture and a hands-on tutorial. Do not worry if you cannot join us in Croatia, as we will also make the material available online after the workshop.
Perhaps the biggest upgrade since our last post is that we changed our boat platform. We now use a slightly smaller boat, called Revere Seahorse, fully owned by the project and that fits in our lab. This allows us to further increase the quality of the dataset as we can now more easily and systematically verify and calibrate sensors and data streams. On the boat we have added two super structures for sensors and a water proof box for our servers, as seen in the picture below.
Another upgrade is that we have changed the lidar setup, as can be seen under Sensor platform. We now use three Ouster lidars, two OS2 and one OS1, for maximum coverage around the boat.
Stay tuned for more updates, and hope to see you in Croatia!
We have now officially launched the Reeds dataset! The world’s largest dataset for benchmarking perception algorithms for robots and vehicles. The idea for this dataset came a few years back when noting that new perception algorithms often showed very similar results, for example towards Kitti, to the ones previously published. Then, when the Waymo set was released back in 2019 we got even more committed to do something, as the new dataset to a large extent contained similar data as the very old Kitti set. We wanted to bring something more interesting to the research community. To not just do the same things over and over, but rather look for more difficult and varied problems, and more honest and interesting benchmarks.
From our activities in the Chalmers Revere lab since 2015 we had gathered quite much experience and software, like our own libcluon and OpenDLV, and the work with creating autonomous machines seemed to contain many of the same problems as creating datasets. Road-vehicles were also getting a bit old, so why not start getting some data from our floating friends? We recently started a good collaboration with colleagues at Research Institutes of Sweden (RISE) and there we happened to have a quite perfect boat involved. So, together with RISE, University of Gothenburg, and the Swedish Maritime Administration we applied for funding from the Swedish Transport Administration and we luckily got the funds to continue.
Now, two years later, we are finally reaching a point when we can start sharing our work and data. Please, if you are interested in joining this journey, to see how the world’s largest dataset for benchmarking robot perception, then please sign up and join us for an exciting future!