Jan 24, 2022
This week's focus: Checkpoint B, which calls for path generation on campus. That's a tall order (and it's topped only by next week's focus on Checkpoint C).
Jim and Egan, champions of this week's behavior planning challenge, already have their code working in simulation:
- Egan has refined our route and path generation, which together produce a collection of general paths for the car to follow.
- Jim has developed what we call the "motion planner," which adjusts the paths from above into detailed curves that will eventually account for obstacles, road rules, even passenger comfort.
The rest of the team has also been hard at work. In no particular order:
- Phu continues to refine his shockingly intricate model of Hail Bopp, right down to the bumper rivets (see below). He's actually creating two versions: A high-poly version for detail and a low-poly version for speed.
- Dylan has formed an Instagram account for us. Give it a follow at instagram.com/nova.utd! He will also be taking pictures and video of our lab work for the next two weeks.
- Rae and Raghav continue to streamline our obstacle detection and visualization features, with Rae forming custom messages and packages to integrate with our camera, and with Raghav building a 3D web application to show the results.
- Connor has been faced with the "sunsetting" of our favorite simulator, SVL. He's now looking at alternatives for us, and will announce his solution soon...
- Kyle is performing the thankless task of map creation and refinement. We need Demo 2's "HD" Lanelet map to be super accurate, surpassing the accuracy of even satelite imagery. Kyle's on the case.
- Josh has added some fancy tools to our build system. Now, writing CMakeLists.txt for our packages is as simple as copy and paste, and Josh's new tools will handle the rest. Anyone who has written a CMakeLists can breathe a sigh of relief. As always, he's also been managing our increasing number of PRs.
- Will (that's me) is adding refinements to our state estimation stack, which we're calling "Atlas." Currently it only uses GPS and IMU data, fused with an Unscented Kalman Filter. To get lane-level accuracy, we'll need additional data from Lidar-based point registration (NDT, ICP, etc.). Adding point registration w/o slowing things down is tricky.
- Avery is re-writing our GPS interface to support "UBX" messages, which encode data critical to our state estimation stack. He's also reading our car's speedometer data from the CAN bus, and preparing to form the steering system required in Checkpoint C.
As you can see, we've got a lot going on! We've even released version 1.2 of Navigator, and we can't wait for 1.3
Most importantly, our project is on track. We have two busy weeks ahead, though, so if you're reading this, keep your fingers crossed. Here are the dates:
- Jan 23 (yesterday): Path and motion planners working in simulation. Done!
- Jan 27: Path and motion planners working on campus (Egan and Jim)
- Requires state estimation, GPS, and speedometer interfaces to be done (Will and Avery)
- Jan 28: Checkpoint B recording finalized (Dylan)
- Feb 3 (less than two weeks): Checkpoint C complete. Self-steering working on campus.
- Minimal map should be finalized (Kyle)
- Steering controller done, tested, refined (Cristian and Nikhil)
- EPAS interface reworked, tested, refined (Avery)
- Obstacles should be detected and visualized (Rae and Raghav)
- Verified with pre-recorded data on Quad (managed by Connor)
- Feb 4: Checkpoint C video complete (Dylan)
Above: The detailed model of our car that Phu is developing.