Hello! I was looking into creating a system in which you can have fully autonomous actions during teleop. This was seen a little bit throughout this last season with teams using approaches like A* and then creating spline trajectories between these points. Im trying to create something more dynamic that can path plan in real time given current velocity, heading, obstacles, and hopefully eventually robots moving through the field as well. I feel like some implementation of RRT would be the most reasonable approach, but I have no idea how you then time parameterize that path once generated such that it fits the constraints of velocity and acceleration. I’ve also figured that you could use RRT but when generating the points you generate them with points only the distance away with which you can move in a 20ms, that creates a whole other problem of numerically integrating your acceleration over time and implementing that WITH rrt seems complicated but potentially doable. Does anyone have knowledge on the subject matter, know any resources, or work in an industry that deals with this? Any help is very much appreciated. THANK YOU!!
I might be totally incorrect here, but doesn’t that just go against the point of teleop?
Our team was planning to use our dynamic path planner to kind of assist the driver during teleop. We were going to have driver 2 select target points for pathfinding while driver one handled overriding in case of unexpected obstacles (we never got around to proper robot detection) or something else that needed their attention. If done right this kind of path generation could definitely help during teleop especially if implemented taking in velocity, heading, and all of that, since it would optimize movements across the field to minimize cycle times.
Hmm. That’s very interesting. I don’t have any advice or resources to offer but I wish you the best of luck 
3044 had an A* solution that worked for it and integrated with an IK system for auto scoring or pickup in the field given any target if you want to learn more the student who lead the development @nab138 should be able to help with in depth knowledge and of the stumbling blocks our team ran into with it and how he resolved them
one of team 100’s students is experimenting with a slightly different algorithm called Generalized Label Correcting, which is a sampling method like RRT, but it samples in control space and simulates the result, so there’s no local planner; the output is simply the control schedule, and another pleasant feature is that it naturally handles moving obstacles.
the paper is here
our java translation of Brian’s C++ code is here:
if you end up playing with this stuff i’d be curious to hear how it goes.
The A* algorithms already developed are beautiful and are an elegant solution to driving between points collision free, hands down amazing, just interested in the more nuanced real world application when there are moving robots on the field. Thank you for the resource!
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.