Aerodesign Payload ·

Payload concept analysis

Built a document of payload options for the team, then sat down as a subteam and analysed concept fragments across motion, interaction, and detection plus localisation. Morphological chart and concept summaries are up next.

What I’m trying to achieve

Develop a document of options for the payload and present it to the team. Then analyse everyone’s options across the three categories: motion, interaction, and detection plus localisation. Get closer to picking a payload design.

Research

Read four external references before the meeting and built a sensing comparison table.

What I read

  • opennav_docking (Open Navigation, sponsored by NVIDIA). Not directly usable (ROS 2 stack, charging-dock semantics), but a reusable insight is: navigate to a staging pose using whatever global localisation you have, then run a vision-control loop for the last metre using AprilTag or similar. Maps to “drive roughly towards the plane, then refine”.
  • SCUTTLE (Texas A&M open-source mobile robot). Complete reference for what a small differential-drive ground robot looks like end to end: hardware CAD, motor controllers, kinematics library, ROS driver. Not the payload itself (heavy, indoor flat-floor assumptions), but a clean example of the layers.
  • Self-Aligning EPM Connector (Wang and Stokes, Edinburgh, 2025). Mechanical connector using electro-permanent magnets. Holds with no continuous power, switches with about 0.3 J. Reported success rates of 59 percent at 0 degrees, 45 percent at 20 degrees… Dunno if we’re going to go with magnets but this was cool.
  • Fast robust peg-in-hole insertion with continuous visual servoing (Haugaard et al., 2020). Supervised learning on synthetic data for precision peg insertion. Sim-to-real generalisation reported. Significant engineering cost to bring up an ML-based stack from scratch.

Sensing comparison table

IR beacon homingLIDAR (2D spinning)UWBCamera + AprilTagSonar (ultrasonic)
What it tells youBearing to the beacon (no distance, no orientation)Distance to nearest surface in each directionRange between modules; pose by trilaterationFull 6-DOF pose of the tag relative to cameraDistance to whatever’s in front of the transducer
Native precision (1 m range)Bearing ~5°, no useful distance~1-3 cm radial, ~1° angular~10 cm typical, ~3 cm tuned~1-2 cm, ~1-2°~1 cm distance, but cone is ~30° wide
What goes on the planeOne IR LED emitterNothing (passive target)1-3 anchor modulesOne printed AprilTagSpeakers (or reflectors)
What goes on the payload4-8 photodiodes + filterSpinning lidar unit1-2 tag modulesCamera + small SBCMics or transducer array
Outdoor sun robustnessWorkable with modulated carrier + 940 nm filter, but the hardest of the fiveFine for spinning units; cheap 1D ToF can struggleUnaffectedFine, manage exposure & glareUnaffected by light; affected by wind
Surface unevenness impactLow (line-of-sight only)Bumpy surface returns can confuse close-range scansNoneLowHigh (bumps echo, especially low-mounted)
Power draw characterVery low (LED pulses, photodiode is passive)Moderate-high, continuous (motor + laser)Low (short pulses)Moderate (camera + SBC)Low
Compute friendlinessBare MCUMCU minimum, SBC comfortableBare MCUNeeds SBC (Pi-class) realisticallyBare MCU
Hardware cost<$10$100-300~$30/module, so $90-150 system~$30 (camera + Pi Zero)<$20
Dev time to working prototype1-2 weeks2-4 weeks2-3 weeks (calibration is the time)1-2 weeks2-3 weeks
Main failure modesSun saturation if filter/modulation skimped; LOS blockedGlass/black/very-shiny returns; rain/dust; mechanical wearMultipath in metallic environments; needs ≥3 anchors for full poseTag occluded, glare, motion blurEchoes off DLZ seams; wind noise; cone too wide for precise alignment

Range stops being a discriminator on a 2.4 m square DLZ. All five sensors comfortably reach the worst-case 3.4 m diagonal. Precision, cost, and outdoor robustness do the actual work of separating them.

Meeting: concept walkthrough

Attendees: Isaac, Aurora, Owen, Leo, Amjad.

Concept gen is due June 13. We have already done a fair amount of external and internal search, so the room had a lot of fragments. Spent the meeting walking through them by category (motion, interaction, detection plus localisation), writing pros and cons as we went. We did not pick anything. The full walkthrough lives in the rolling meeting minutes document for future reference. Post-meeting action is to move all of it into a morphological chart so we can start building combinations.

Driving questions

  • How many motors can we stuff on the payload?
  • How configurable is the bottom of the plane?
  • Do we want to give up on the flight time bonus in favour of making sure we get the payload delivered and captured perfectly (take our time)?

Next

  • Move every concept fragment into the morphological chart (image to be drawn).