What we are specifically interested in is for our quad-copter to be able to detect and maneuver to avoid (maneuverability is based on our custom algorithms) objects flying towards it with great speeds. A paintball gun projectile for example. Regarding the coverage of all possible directions, we are not concerned whether or not it will be achieved with one or many sensors. What we are interested in is if a sensor that can do that exists and is suitable to be mounted on-board a drone.
$\begingroup$
$\endgroup$
7
-
1$\begingroup$ LOL. You're asking for a sensor that can sense a paintball pellet within 25 m for less than $1? I'm hoping that somebody has such a sensor because I'd buy a bunch of them. None of the sensors that I am familiar with can sense something that small and fast at 25 m. None of them can do it at one meter. The closest one that I can think of is a very fast camera. $\endgroup$– NomadMakerJun 12, 2018 at 1:05
-
1$\begingroup$ @NomadMaker, that number may be a thousand dollars $\endgroup$– jsotolaJun 12, 2018 at 1:44
-
$\begingroup$ @Panteleimon Stamatakis Have you considered dynamic vision sensors? (event based cameras). While as of now, they are above 1000 USD, It is rumored that Samsung is going to come up with cheaper event cameras this year. $\endgroup$– HighVoltageJun 12, 2018 at 4:09
-
$\begingroup$ Sorry, I'm used to the comma as the separator for the thousands. I like the dynamic vision sensors. They are considerably above \$1,000 USD, but all technology comes down in price, eventually. I'm wondering if a fast enough traditional camera plus perhaps an FPGA could be bought for under \$1,000? Looking for a fast paint-ball pellet at 25 meters would require a lot of computer power. $\endgroup$– NomadMakerJun 12, 2018 at 6:01
-
$\begingroup$ @NomadMaker Yes, I was referring to one thousand dollars. So radar or lidar is out of the question for this kind of stuff then? As far as compute is concerned, we are opting for the nVidia Jetson tx2. Do you think it is under-powered for this particular task? $\endgroup$– Panteleimon StamatakisJun 12, 2018 at 7:13
|
Show 2 more comments
1 Answer
$\begingroup$
$\endgroup$
3
Consider a pellet going 100m/s, it will cover 25m in 0.25sec, let's say your drone is 0.5m in size, and detection time is zero, you still need to achieve 16m/s2 (1.6G) acceleration instantly to get out of the harms way -- which is a bit unrealistic for the contemporary drones.
Take a 60FOV camera, with 4k resolution, this will give you about 2000 pixels vertically (or horizontally), which @25m become about 12.5mm large (one pixel), which is not enough for the reliable detection. 30FOV camera might work with 6mm pixels @ 25m, but you'll need plenty of them to cover your surroundings. And processing all these 4k images is not going to be cheap or fast.
You cannot detect a fast flying object @25m unless you use high velocity cameras (upward of 100fps), which again drives the processing cost higher.
IMHO, your best bet would be to go either in the direction of creating a large reinforced drone with high redundancy that does not care about pellets and can keep flying after the hit, or creating a fleet of match-box sized randomly moving drones that work as a swarm and don't care if there are a few casualties. Check out this: https://www.youtube.com/watch?v=9CO6M2HsoIA
-
$\begingroup$ Really informative answer thank you. The reason why I noted these kind of metrics is due to the relative velocity. The drone might not be still and have targets incoming at it, but might be travelling really fast and try to avoid object in real time by maneuvering (we are opting for a GA approach trained in simulation) and then implementing it through calibration based on the motors. Nevertheless thank you for taking the time to write this, any kind of insight is deeply appreciated. $\endgroup$ Jun 14, 2018 at 9:35
-
$\begingroup$ @PanteleimonStamatakis I would not undertake such a project. With the current drone technology they have troubles flying around large static objects. From 25m I'll take it down with my bare hands just by throwing gravel with a relatively large shovel =) $\endgroup$– lenikJun 14, 2018 at 9:42
-
$\begingroup$ The tech is fine. The software is bad, due to the fact that enthusiasts who do this kind of stuff have no idea regarding complexity and optimization. $\endgroup$ Jun 15, 2018 at 6:21