Some non-technical reasoning I believe in strong decouplig of the autopilot and the companion (this doesn't mean they can't live within the same box). However plug&play is a great plus, as this means that I (final user) can buy an Iris / Pixhawk / whatever and in a second moment I can add the companion as a second layer of computation, while all the underlying settings remain the same. This mantains a very attractive learning curve. In a third moment I can even switch to a more powerful companion as new needs / new supercomputers arrives.
In this group we've been dealing with companions fore a long time. Now @Randy opened a repo called "Companion" https://github.com/diydrones/companion. A very big thank you for this! I invite everyone to contribute.
SUGGESTION ABOUT COMPANION REPO:
- COMMON (this is a directory with common scripts. @Randy maybe also a simplified version of balloon fineder could fit here: colour_finder_web.py ). I'll put also VPN scripts, gstreamer (@Bernt please contribute!!) I can also contribute with 4Gmetry .py scripts.
- UBUNTU / ODROID XU4 this is what we're doing here at 4Gmetry until now.
- UBUNTU / JETSON TK1 this is the supercomputer. As the main usage will be computer vision, I think that we should give a look at http://elinux.org/Jetson/Computer_Vision_Performance as suggested by @Bill_Bonney
- n OS / BOARD combinations
- PERFORMANCE this includes performance tests across the above combinations.
What do you think?