Plan:
Integration by Priority:
P0: Aircraft should have a functional frame, be able to fly, grabber operation, static gimbal and be able to send video to ground. COTS FC and PDB.
What’s Left: Sending video over openhd (CV), grabber mounted and operating, operation from Pixhawk PWM output (FW) [this would require significant knowledge transfer from Andy, or Andy would have to go, gimbal mounted onto aircraft and operating.
Tasks we need to focus on:
Grabber can be mounted but it still has current draw that’s too high, rectifying this issue would require us to up the voltage w/ a step down converter or buck converter from 24V to 12V, but that would use a lot of space. We could also try using a linear actuator for the grabber, choose based on weight, current draw, etc. Decision will be made NLT next week. A COTS PDB would also have a 12V output, and that would be the easiest option.
Path Forward: Mech will continue testing a linear actuator option but unless there’s a very high degree of confidence from Mech, we will go with either a COTS PDB or a buck converter from EE.
Gimbal has small issues with tolerance and sizing, holes are being resized for M2, we’re just having it reprinted this weekend and reassembled next week. Frame mounting is currently incorrect, can be bodged for flight testing, but the next frame will have this fixed.
Path Forward: Reprint and mount by next week.
P1: Finalized frame, flying off ZP with human control only + P1
Frame: Current iteration is done, next iteration will be done by the week of the 29th and we want to flight test it then, current iteration will just test grabber integration.
ZP Controls + Peripherals (Grabber): Start by trying out the MPU6050, see if it works with our setup, if it works, then we know that the Zeropilot board is fine, continue testing BMX IMU. Else, try different ZP board. Next few days, we’ll validate with SPI driver to see if it’s our code or something external to our code, and then after that we’ll have Aidan in on Monday evening to work through BMX IMU issues. After IMU is resolved, we just need to PID tune and test. PWM will be replaced with DSHOT over the next few weeks, need to scope out work to get DSHOT to work w servos. Note that Arduino is also available.
Path Forward: Focus on getting IMU resolved, by next Tuesday we should have an idea of if we can continue using the current architecture for IMU or switch to getting IMU from MCU chip. In the meantime, continue w PID tuning once MPU6050 is validated. PID tuning should be done by the end of the month.
P2: P1 + Autopilot + Full communication for Jetson - FW - Ground + Detection
Tasks we need to focus on:
Autopilot: PR should be in by end of the week and we’ll have a better estimation of time by the end of the week but roughly could be validated by the end of the month. VN300 drivers are still in progress but should be done by end of the week.
Path Forward: Wait for build issues to be resolved for VN300 and PR for PM to be in before we can fully determine scope.
Comms: Jetson still needs to be setup, Tensorflow been an issue, let’s get the Jetson setup by this Thursday by dropping Tensorflow and doing the rest of the setup. After that, FW-CV comms will be the first to be validated, we’ve confirmed that the FW end is likely alright so now any furhter issues are going to come from the CV side. Ground Comms, still need to set up the XBees on the Ground, Jetson and FW. Ground and FW setup should be done this week, we’ll scope Jetson setup after Thursday.
Path Forward: Get Jetson setup on Thursday, scope effort for XBee on Jetson, setup XBee on ground and FW this week. If ready, we can run FW-CV comms test this weekend, and we should run Ground - CV tests NLT end of the month. Ground - FW will be run this week.
Detection: Detection training is not going great as the accuracy is low with mixed datasets. If mixed datasets fails, we’ll stick to using one dataset and/or the test flight dataset. Model deployment is complete.
Path Forward: If video transmission works and we get training complete, detection is complete. Continue one more round of mixed datasets, if it fails, fall back to one dataset and/or the test flight dataset and deploy models ASAP for testing.