...

https://uofwaterloo-my.sharepoint.com/:f:/g/personal/uwarg_uwaterloo_ca/EnY_jjQ-qNxBvzZG8dXqsmcBWxHpaoz841fN5Z3HShZSRA?e=zyI504

Actual Timeline Recap

...

9:05 - brief

1000 - Left the bay

1015- Arrived Wrest RC

1032 - Pegasus Takeoffs [50.BIN]

1042 - Pegasus GPS test [

1047 - Pegasus OpFlow test

1049 - Houston Proximity detection (1 stab, 1 loit)

1059 - Autotune on Pegasus [

1145 - Drone picture

+2 more flight on houston - battery swap.

Batteries

1 set of battery for pegasus (4 batteries +) - total FT unknown

4 batteries for Houston - total FT unknown

• almost all batteries on houston around 14% (guessing 10.6v)

Green/Yellow/Red

Analysis

Pegasus Op-Flow Calibration [0050.bin]

Took off + op flow calibration worked great

• FlowX and FlowY scalars set to -158 and -152.

• Took about ~ 10 seconds to calibrate, required pitch/roll inputs.

EKF failure when transitioning from GPS to non-GPS modes → Entered failsafe and landed.

• can takeoff in GPS mode and then switch to EKF SOURCE Middle (opflow only)

  • can maintain stable flight / position controlled flight for ~ 10 seconsd before EKF failsafe

• Had a bit of drifting in xy plane in loiter (not maintaining a very good position hold). Could be due to GPS error (but had opflow?).

  • occured in both GPS and OpFlow modes.

Notes/ next steps:

• determine root cause

  • See if error is re-producable

• Do not fly over people/buildings until source/cause is known (potential to enter EKF failsafe upon gps disconnection?)

• Attempt to determine solutionss

  • Potentially re-configuring EKF sources / blending

  • Potentially re-calibrate accelerometer

  • Potentially examine gps-nongps transitions in more details

Houston proximity

• Maintained 2m distance (set distance) when flying in loiter towards obstacles.

  • Worked in all directions.

  • Worked the best from the front, not significantly different in all axis.

  • Harder to tell if input commanded was aligned.

• If you had more momentum approaching the wall, the drone would over-compensate and “bounce” away from the wall.

  • If full throttle, it would move back to 2m.

  • If backed off the throttle, maybe a bit of oscillation.

Pegasus Autotune

• Pegasus Yaw had significant amounts of overshoot

  • Fixed after re running autotune

Houston Auto mission

• Took multiple attempts to path around the object.

• Saw the house, was trying to go around? Seemed like it couldn’t find the place…

  • there were 3 meters high bushes around the house'

Some failure points

• 4 meters was too tall for the shed/house we were using as a wall obstacle

• auto mission went close to a pole, resulted in pilot disarming the drone

...

Time	Activity	Notes
0700	Battery Charging Began
0905	Briefing began
0945	Battery Charging finished
1000	Left the bay	3 vehicles
1015	Arrived WrestRC	Field was covered in slush/pooled water. Only one dry track
1032-1034	Pegasus OpFlow Calibration (1-1)	Log file 50.BIN
1042-1043	Pegasus Loiter - GPS Handling (1-X)	Log file 51.BIN
1047-1047	Pegasus Loiter - OpFlow Handling (1-X)	Log file 51.BIN
1051	Houston proximity test (3-1)	Log file 70.BIN Batt #1
1059	Pegasus Autotune (X-X)	Log file 53.BIN
1110-1112	Houston Auto Mission (3-3) Attempt #1.	Log file 71.BIN Unsuccessful - spun in circles trying to avoid the obstacle Batt #2
11:15:15-11:15:45	Houston Auto Mission (3-3) Attempt #2	Log file 71.BIN Unsuccessful - disarmed prematurely. Batt #2
11:45	multi-drone photo
11:50:33 - 11:54:50	Houston Auto Mission (3-3) Attempt #3	Log file 73.BIN Unsuccessful - Altitude too high
11:57:20 - 1159:51	Houston Auto Mission (3-3) Attempt #4	Log file 73.BIN Unsuccessful - Got stuck in turn-around due to embankment being detected as an obstacle.
12:04:15 - 12:14:27	Houston Auto Mission (3-3) Attempt #5	Log file 74.BIN Successful - passed obstacle after 4 attempts. Had a very hard unscheduled, uncontrolled, rapid descent.

Tip

OPFlow was calibrated, EKF Transitions were tested (Card #1 Complete ✅ ) Pegasus not damaged. Houston Obstacle Avoidance verified both in Loiter and in Auto missions (Card #3 Complete ✅ ) Landing gear & houston was kept in-tact despite multiple hard landings / failsafes Lidar not broken and continues to function RFD900x was swapped from Pegasus to Houston very quickly. ELRS did not have the necessary air-data rate. Pink controller worked Models were copied day of Media gathered Pilot practice is good We left on time, didn’t forget anything

Note

Obstacle avoidance took multiple attempts to pass obsttacle (Card #3) Pegasus vibrating large amounts pre-takeoff on arm. Pegasus controls were not great Huge amounts of Yaw overhsoot → disappeared after autotune. GPS Loiter not very accurate - huge amounts of slip when stick-centered before drone attempts to stop motion. Baro/Altitude solution not very accurate - noticed huge altitude drops when in transaltional motion. Automatic recovery when stopped. Houston GPS Loiter not very accurate Multiple meters of drift, no optical flow. Arm switch on pink controller bit close to the mode switch (in Megan Spee opinion) Models were copied from Blue controller to pink Controller before flight test & without pilot checks. didn’t note down total flight times or battery voltages post-flight Total flight times recorded on controller, but not noted. Takeoff times not recorded. Battery voltages not recorded. Auxiliary video/FPV drones around works for media but was distracting/ interfered with the tests being within yelling range helped: could tell people to back off TBS Crossfire consistently transmitting at 500-1000mW. Likely due to RFD900 + TBS Crossfire being on the same frequency and very near each other.

Warning

Tracking antenna was not tested (Card #2) Was not ready before the flight test Gemini was not tested Was barely ready before the flight test, ran out of battery on Houston. Houston experienced a rapid un-controlled descent from ~ 40m Likely a battery FS Houston batteries ran down to 7.6v under load - 10.4 V after recovery Pegasus EKF failures when doing GPS/Non-GPS transitions using RC7 as 90. LTE connection was dropped at some point (multi-drone photo). could not fly after this.

Analysis

All log files are available in onedrive, see section above. On Average, ELRS Performance was awesome, while TBS Crossfire was typically transmitting at close to max power, likely due to interference from the RFD900 on the same frequency.

Pegasus Flight(s)

All Pegasus missions were run using LTE, which later failed to connect when attempting the multi-drone photo. It was impossible to takeoff in EKFPosSource MIDDLE, and we only took off in EKFPosSourceLow (GPS mode). Initial YAW tuning on pegasus was quite bad (lots of overshoot), but auto-tuning was able to correct this.

Pegasus did have issues maintaining altitude during transition - it tended to drop ~1m in altitude in translational movement, and had a tendency to level off + continue sliding once the stick returned to center.

The deadzone for takeoff + landing throttle on Pegasus is also significantly larger (read: harsher landing + takeoff) compared to Vanny or Icarus.

Autotune

...

FlowCAL + GPS/non-gps transition

...

• Flow calibration worked as expected

...

EKF Failures:

• 10:34:18 : EKFPosSource set to MIDDLE

• 10:34:33 : EKF3 IMU0 stopped aiding

• 10:34:33 : EKF3 IMU1 stopped aiding

• 10:34:33 : EKF3 lane switch 1

• 10:34:33 : EKF Primary Changed: 1

• 10:34:33 EKF variance

• 10:34:33 : EKF Failsafe: changed to land

...

• once landed, all failsafes cleared & position was set back to middle.

• Position estimate / IMU2 was still able to provide level landing assistance.

Loiter GPS aided flight

...

• estimated position looks good

• No harsh events

Non-gps aided loiter attempt

...

• took-off around 10:47:12 - landed around 10:47:42

• Flight path looks good, position estimates were continually given.

• EKF IMU stopped aiding messages were given at 10:47:34

• EKF Failsafe at 10:47:35.

• Had a few seconds of non-gps aided flight? Not sure why the failsafe occured.

Houston Auto Mission(s)

Ran a total of 5 Auto missions attempts, eventually successful in pathing around the object.

4 Batteries were used, and OA distances were adjusted as follows during the FT:

• OA_Type: Dijkstra with BendyRuler → BendyRuler

• OA_MARGIN_MAX: 5m

• OA_BR_LOOKAHEAD: 15m → 10m

• AVOID_ACCEL_MAX: 0

• AVOID_BEHAVE: 1

• AVOID_DIST_MAX: 5m

• AVOID_MARGIN: 2m

Attempt #1

...

• Houston was successful in not colliding with an object in the direct flight path.

• Houston was unable to successfully path around the object and loitered in circles.

• Mission was ended by the PIC.

Attempt #2

...

• Path was updated to move houston further away from the building on waypoints #3 and #4.

• Houston successfully avoided a light-post, but was disarmed pre-maturely when PIC wanted to switch to Stab (SB) and accidentally hit Disarm (SA). The drone dropped from an altitude of around 8 meters, no damage.

...

• Orange is 45 degree sector, and green is 90 degre sector.

• Notice that the MIN from 45 & 90 is 15 meters (This was the configured OA_BR_LOOKAHEAD).

  • This may not be correct - the physical distance from the pole was significantly less (visual estimate 1-2m upper end)

Attempt #3

...

• Houston over-flew the building. Auto mission altitude was set to 5m, and was too high.

Attempt #4

...

• HSTN got stuck since it detected the embankment as an obj.

...

Attempt #5

...

• Houston took multiple attempts to navigate around the obstacle.

• Possible to see proximity sensors pinging off the obstacle and reporting closer values as houston navigates around?

...

Next-Steps

• Update & clarify transmitter mappings, common channels, etc. for all drones.

  • arm/disarm sequences, mode channels, aux functions, telemetry logging, etc.

• Test gemini, tracking-antenna

• Remember to note down takeoff & landing times of all drones (at least total FT)

• Remember to have videos of drones during FT’s.

• Evaluate Pegasus tune.

• Evaluate operation under GPS failures.

• Come up with plan to optimize obstacle detection

Action items for next flight test

...