Mini itx PC inside an UAV (probably an hexacopter).

[Pawan Sharma]

Guys! I Need help, So i am planning to make a UAV which would be flying at the speed > 60 Kmph (38 mph). I am planning to use ZED camera for detecting obstacles and image processing. Spec sheet of ZED cam specifies that it need a processor with core clock > 2.3 GHz and cuda supported "NVIDIA" GPU with compute compatibility > 3.0. So i am planning to use a low power mini ITX PC onboard powered by AMD Athlon 300GE (Because of its low TDP and clock speed of 3.4 GHz base) with a nvidia GPU (probably GTX1650). Assuming that we have a way around Weight and power consumption. Should i worry about  endurance of these components at this speed? Should i worry about static electricity damaging the components?

 

We were initially planning to use nvidia's jetson boards but they can achieve max of 1900MHz (after overclocking).

 

Please Guys I want your suggestion.

[Dumb_Ass]

Hey, can u be more specific with what type of uav platform (chassis) will be used in a bit more detail? 
Have u consider using jetson tx2? Or even  dropping idea of using ZED completely and going for something like RPi4 or Panda and stereo expansion card for cameras? 
Let’s be honest there is not much to avoid in mid air and u not going to fly between branches or power lines anyway. 
A bit more detail will be much appreciated.

[Hackentosher]

I would strongly suggest using a singleboard computer such as an Nvidia Jetson Nano, Up Squared, or some kind of Pi. Power alone is going to be a pain in the ass if you want to build an itx system and run it off a battery. Also remember that clock speed =/= processing speed, so I would suggest looking around and comparing benchmarks not clock speeds.

[Pawan Sharma]

8 hours ago, Dumb_Ass said:

Hey, can u be more specific with what type of uav platform (chassis) will be used in a bit more detail? 
Have u consider using jetson tx2? Or even  dropping idea of using ZED completely and going for something like RPi4 or Panda and stereo expansion card for cameras? 
Let’s be honest there is not much to avoid in mid air and u not going to fly between branches or power lines anyway. 
A bit more detail will be much appreciated.

Hey! Thanks for replying.

We are designing this for a competition (Sorry ;Can't disclose the name of the Competition), so yeah obstacles willl be planted purposefully and also they would be in motion. So we really need powerful depth camera and according to our knowledge (Which may be limited) this is the only way. We thought of using RP-Lidars at first but they are not a good option in bright environment. Sonar also can't be used due to interference from the sound of the motors. RADAR is also not an option because they need huge power to operate and also are hard to build.

 

We already tested our prototype with jetson TX2. It was not as efficient as we wanted. It took a second to process a single frame which is slow for our needs.

 

As of type of UAV its an coaxial Octacopter. It needs to travel 6 KM (path is Elliptical), detect and avoid obstacles,  Replace a module fitted with electromagnets (kept on an tower like structure ,that is it should be capable of hovering) in or under 9 minutes. We thought of going with VTOL structure at first but octacopter is enough to fulfill that need and is easier to work with.

 

Any suggestions?  

[Pawan Sharma]

7 hours ago, Hackentosher said:

I would strongly suggest using a singleboard computer such as an Nvidia Jetson Nano, Up Squared, or some kind of Pi. Power alone is going to be a pain in the ass if you want to build an itx system and run it off a battery. Also remember that clock speed =/= processing speed, so I would suggest looking around and comparing benchmarks not clock speeds.

Hey, Thanks for replying.

 

We already have tested Jetson TX2's and they are not fast enough to process data that we need to avoid obstacles at the speed > 60kmph. And Jetson was our only option in singleboard computers because ZED needs nvidia graphics to run and that too it need the ones with compute compatibility > 3.0 and CUDA Tech. Also If have anything in mind that's better than ZED for obstacle avoidance ,it would be greatly appreciated.

[Dumb_Ass]

2 hours ago, Pawan Sharma said:

Hey! Thanks for replying.

We are designing this for a competition (Sorry ;Can't disclose the name of the Competition), so yeah obstacles willl be planted purposefully and also they would be in motion. So we really need powerful depth camera and according to our knowledge (Which may be limited) this is the only way. We thought of using RP-Lidars at first but they are not a good option in bright environment. Sonar also can't be used due to interference from the sound of the motors. RADAR is also not an option because they need huge power to operate and also are hard to build.

 

We already tested our prototype with jetson TX2. It was not as efficient as we wanted. It took a second to process a single frame which is slow for our needs.

 

As of type of UAV its an coaxial Octacopter. It needs to travel 6 KM (path is Elliptical), detect and avoid obstacles,  Replace a module fitted with electromagnets (kept on an tower like structure ,that is it should be capable of hovering) in or under 9 minutes. We thought of going with VTOL structure at first but octacopter is enough to fulfill that need and is easier to work with.

 

Any suggestions?  

Ok,

So, what was the resolution? Colour scheme (canada eh)? Did u optimized code? 

The second for one frame is way off even from specs of jetson, 5 fps is what it suppose to give judging by specs.