I have seen a post about this from a few years back but I wanted to re-ask the question now considering the game is different this year and things may have changed. Is the vision sensor worth using in the autonomous? My team has one but I have never used it before, and reviews online seem to be mixed about it’s reliability. I was thinking that it could be used to detect the triballs this year, which it might work pretty well for because there are no other game elements with a similar shape or colour this year.
If you have perfect odometry and great use of distance sensors, you can live without vision.
However if you find your robot misses a triball intake, during your auton maneuver… vision might be your answer. You can use the vision sensor to feed your driving PID while intaking, and it could, with careful engineering and programing, fix intake inaccuracies and create a reactive and dynamic auton routine.
Also, shape is not viable for the vision sensor. They do not return any information about shape.
I think it’s really just up to personal opinion. They are often really useful in certain games and can be very useful in Over Under with the triballs. They are extremely useful when tuned well but is very time consuming to test and tune both before and at competitions. This year, because the triballs are on the ground, items outside the fields won’t be as big of a distraction for the vision sensors (e.g. Sponsorship Banners, Signs etc.). Personally, I find that the need for tuning and risk of distractions that can put off the vision sensor makes it difficult to work with sometimes, which is why I don’t use them anymore. But i’ve seen teams do well with and without a Vision Sensor. I encourage you to do some testing with it so you can determine whether you like them or not.
I have experimented with them and have found them to be too difficult to tune to make them worth the time.
The best use of the vision sensor is to identify objects that are on the ground near the robot. This could be very useful in autonomous to correct for objects that are misplaced or have been moved a bit by other robots. So, yes there is a possibility of value in auton. That said, most teams overcome the slightly misplaced object by the mechanical design of their intake.
What the vision sensor is not great at is aiming at flags in turning point or goals in spin-up – high objects far away where room lighting and background conspire against the very low pixel dimension of the sensor.
2 Likes
Having multiple vision sensors, and then for-looping through a selection of various signatures for which has optimal metrics can help with this. While you cannot just average the data from those images, you can do some cool logic to determine the likely best representation, and then continue to use that signature for the low latency PID stuff while you continue to scan the other signatures for a possibly better one for that moment/angle/shadow/etc.
Thanks for all the replies guys. I briefly tried configuring the Vision sensor with the video that shows up on my computer and it does seem to need tuning at every slight change in lighting. Because of the tuning issues (don’t want to risk my auton working because the sensor is not tuned correctly in a tournament), I will do as @ApertureLabs said and focus on making my odometry better for now, probably will revisit the sensor later.
If you have time to cook over holiday break, I would recommend trying multiple signatures for each color, and then take the one that results in the best metrics, whatever that might be. As hard as it is to take one picture and pull values, taking three and comparing is almost trivial. There are lots of ways to optimize some type of constant auto calibration, but its largely unexplored and definitely not included in any of the main public libraries that are being passed around.
IMUs have come a long way from accel/gyro given that they constantly auto-calibrate now. It would be neat to see if in the future we get some neat vision tools.
This topic was automatically closed 365 days after the last reply. New replies are no longer allowed.