I have a HereFlow and a Lightware LW20/C that I’d like to test again - I haven’t used them for over a year - past when the latest optical flow improvements were incorporated.
Before I get started, I’d like to understand what to expect - either with or without a GPS failure.
Of course the LW20/C will provide terrain following.
If the GPS is operating normally, will the HereFlow add any additional capability to the autopilot?
If the GPS experiences a glitch, or simply fails, how much can I rely on the optical flow capabilities to navigate?
As I recall, the HereFlow sensor is pretty small - 16x16 pixels if I remember correctly. I don’t know if the autopilot uses that input to build a “map” that can be retraced.
Speaking of “retraced” - I assume that any navigational ability of an optical flow device relies on it travelling across terrain that has been previously traversed. Is that a valid assumption?
And how much terrain in the optical flow sensor’s map is retained?
I’ll try to test the HereFlow’s performance and see if it matches the expectations based on this post’s responses.
Optical flow doesn’t do global contour mapping, it only provides information about apparent angular velocities.
It will likely constrain position drift so EKF won’t happily fly away in a case of GPS loss but position accuracy will degrade with time and maneuvers.
From your description, it seems like the navigational abilities of Optical Flow are similar in some ways to dead reckoning.
Interesting that you bring up EKF. Typically if a GPS fails or gets too many glitches a EKF failsafe will occur.
How does optical flow feed into the EKF3 algorithm in this instance? Will EKF3 status remain good only as long as the vehicle remains in place?
I recall seeing a video in the wiki a few years ago (before the latest improvements to optical flow in the firmware) where optical flow can navigate a RTL.
I’m wondering to what extent such a RTL is possible.
Similarly - if the vehicle is on an auto-mission, will optical flow allow identification of waypoints if the GPS is in operative?
IiRC EKF failsafe occurs if it loses confidence in its state. GPS glitches make situation worse as they cause EKF to get contradicting inputs. With no GPS to validate and correct position estimate it also loses confidence. Here optical flow helps by providing velocity estimate thus going from integrating twice to once so errors and uncertainty don’t increase as fast.
Optical flow works on the move but not necessarily over moving objects.
It improves position estimation without GPS so it might be possible to keep good enough position for long enough to do reasonable RTL over appropriate terrain.
If you want to have positioning without GPS you need to go for full blown visual navigation preferably with map matching or loop closure (if using SLAM).
In order to use optical flow in a mission critical implementation, the performance criteria needs to be known so they can be tested. Without this - an optical flow device is just a novelty.
I appreciate your input. I need greater specifics to be able to do meaningful testing.
The information to establish appropriate expectations might not exist. I’m not familiar with even anecdotal reports of optical flow applications.
As several manufacturers offer optical flow devices, there appears to be some market for them. I’d really like to know more about the practical applications in use.