2024-09-21 Autonomy Roadmap
- Maxwell Lou
- Jane Zeng
- Balaji Leninrajan
- Ashish Agrahari
Attendance
@Maxwell Lou
@Balaji Leninrajan
@Ashish Agrahari
@Jane Zeng
@Aaron Wang
@Mihir Gupta
Discussion topics
Oct 3, 2024 Autonomy system has been ground tested on Pegasus 2
Oct 5, 2024 Fly Autonomy on Pegasus 2, run Geolocation and maybe clustering
Oct 26, 2024 Fly task 1 code for the first time
Nov 2, 2024 Fly task 2 (might be tight)
May 9, 2025 Competition
TASK 1: Autonomous take-off / landing, but manual flying in between? (seek clarification)
Front camera only required to get to fire zone if autonomous (finding red balloons) (if not autonomous, pilots already have front camera)
1. Take off autonomously to a certain height
2. Pilot fly to balloons
3. Start search pattern and detection of IR emitters. At the same time, pilots can monitor the video feed to find the “source of fire”
Need to decide what happens when source of fire is found. Most likely solution is to do geolocation groundside
Other solutions involve either pausing the mission and going ontop of source to get coordinates and then resuming mission, or coming back to coordinates once search pattern is done
4. Pilots fly back / RTL
5. Auto-landing script
TASK 2: Only run autonomous water delivery once (last run)
Pathing (all locations are given beforehand)
Decide which buckets to go to
Request pilots to leave closest bucket for autonomous
Only autonomy when within 5m sphere of water buckets/unload buckets
Request another camera, 1 IR (for task 1, directly down), 1 RGB (directly down for task 2 finding buckets + task 1 finding source of fire for pilots)
Or it might be possible to only use 1 camera. For task 1, make it have the IR filter. For task 2, take off IR filter and use as regular camera
Depends on mech’s implementation of picking up water/releasing it, might be offloaded to EFS
Geolocation - Top priority Oct 26, 2024
1. Using geolocation to get location of IR emitters
2. Fly right on top of IR sensor, then just read GPS data? Centering above the IR emitter
3. Work on both solutions at the same time
Detection stuff (combined with ML) Oct 26, 2024
Can’t fly too high - flight test it to see how high we want to fly (flight test, starting next term after Geolocation and Detection is done)
1. Use ML model to find “brightspots” (new data collection and more)
2. OpenCV to find “brightspots” ← most likely solution
Cluster Estimation - Top priority Oct 26, 2024
1. Take in large number of data points/locations from Geolocation, and group them
Autonomous water task
Loading, unloading
Precise coordinates, or precise controls to center drone on water
Use CV camera, for edge detection of buckets for loading and unloading (as GPS coordinates might not be precise enough)
Airside vs groundside compute
We are doing AIRSIDE compute for this competition
Also, video feed will most likely sent from drone to groundstation
Search pattern
Search within radius of 100m from “sign of fire” (tower of balloons) Oct 26, 2024
1. Go to sign of fire as center (mark coordinates down manually)
2. Fly spiral/concentric rings (either manually or have someone input it into mission planner or have a script ready)
Generate this path given center, radius, drone’s view width
Eventually merge Pathing into “airside”
Start with searching manually, eventually get to autonomous search pattern
KML Oct 26, 2024
Record the coordinates from detection, then translate into KML file
Do read CONOPS for formatting
IR Camera
Detection - Fires
Detection - Source (940nm IR emitters)
Possibly Detection - Fire sign (red balloons)
Final Verdicts/roadmap