After some full speed results, the network once again fell over. It seems too bandwidth intensive to work during other network activity. Beacons needed to be lowered from 100Hz to 50Hz. The ground camera sometimes loses its connection. A backlog of frames occurs, causing the turret to lag & get lost.
For a while, the left eye locked up erratically. Then it started working, for no reason. They're the usual glitches & failures of a new hardware platform. It'll be a long time before it works perfectly. A switch to USB is likely, the longer the wireless problems persist.
The 1st flights with the vision system were bad. Y seemed more stable than X. The vision system may be just fast enough at measuring distance to stabilize Y, but too slow at measuring X. Having the 2nd camera looking at it from the X direction instead of relying on parallax would work better. Then you have a clone of a Vicon system, with multiple cameras in different parts of the room.
The Blade CX2 may be too unstable for any previous navigation system to keep in a small enough area. It might work outside, but what's the point.
It's disappointing to not have the same success as shrediquette, which uses a ceiling mounted webcam to successfully fly it in a very small area, with yaw changes. That uses IR leds that are always on & an IR hacked webcam. The tricopter is probably much more stable than the Blade. It also had a lot more time behind it than 1 month.