Just a quick update, 2020-11-17T19:00:00Z, DARPA is releasing a webcast of the Subterranean Challenge Cave Circuit Virtual Competition. Due to the complexities related to the COVID-19 pandemic, the Cave Circuit has gone fully virtual and DARPA and the Open Robotics crew have been working hard to expand the features available in this round including some interesting video capture capabilities. The whole process itself is really interesting as Open Robotics has been hard at work digitizing the cave environment, modeling some of the Systems Competition robots and their capabilities, creating simulation tutorials, and executing the simulations. This lightning talk from ROS World gives a quick teaser for the event.
I would highly encourage you to tune in and watch the event tomorrow, and take a look at the SubT Hello World tutorials to get more familiar with the software that makes it happen.
I’m really curious on how robust / successful the navigation stack is in the non-flat environment. I took a glance over the tutorial on the subject and hoped that perhaps more information could be provided in how that is working out for folks. It seems reasonable for a first-order estimate. If it happens to work out un-reasonably well for such a simple idea, perhaps we should integrate it into the Nav2 stack directly for the short term until we have a gradient estimation solution.
On that subject, there’s a reference to a fake frame, we typically call that base_footprint in navigation-speak and that’s a totally legitimate thing to do. I might suggest updating the language to use base_footprint since that’s a common step and other resources are available for search under that name. The second side on filtering, I’m curious how robust that is and if there’s any places in the track that it breaks down for for the above reasons.
Agreed it would be nice to know how many teams used the nav stack because I did see a couple of videos with custom planners in 3D with octomaps, grid_maps etc but those approaches seem much more complicated. The suggested changes seem to be similar to those in the RTAB-Map Stereo Outdoor Navigation Tutorial, where base_footprint and filtered obstacle point clouds are used. While usable with these changes, I imagine the nav stack might fail in multi-level situations.
Regarding outdoor navigation, we are currently working on a real 3D surface navigation integrated with Move Base Flex, see https://github.com/uos/mesh_navigation.
In case you’re interested, the Co-Star team goes over their path planning approach/NeBula system briefly in this video. There are also more details in this paper.
