Exploring air entrainment: a lobed mixer in place of the stock exhaust cone. This one is printed for a fit check. The real one will be printed in Inconel to survive 650掳C.
6
Which way a wingtip propeller spins can cut a wing's drag.
Every wing leaves a swirl of air at each tip, and that swirl is wasted energy. Put the propeller at the tip and spin it against the swirl, and part of that energy comes back. In TU Delft wind tunnel tests this cut drag by about 15% compared with a propeller mounted further in.
Small electric motors make tip mounting practical. NASA moved the X-57's cruise motors to the wingtips for exactly this reason.
馃 Made with AI
27
Project Update: the rudders now use a TPU hinge printed in place with the surface. It's flexible, it can't come loose, and it adds zero hardware. It's working really well so far.
19
Project Update: 12S1P battery pack for the pods, built from 12 Molicel P50B 21700 cells. That's 5Ah at 43.2V nominal (about 216Wh) and 50.4V fully charged, with a 60A peak discharge.
16
Project Update: three tail fins off the printer and on the table today. Yellow skins, black root fittings. Next step is fitting them to the airframe.
13
Project Update.
Landing fin v1 and v2 side by side.
v2 gets a larger control surface, and the servo is now built into the fin.
Fewer parts hanging outside means a cleaner fin and more control authority from the same footprint.
When four drones share a cable-hung load, the box can stay on path while some cables go slack and others overload.
AFD-CAMLs (TU Delft) picks a feasible force split during planning, then nudges each drone from onboard tension estimates so the load stays shared. Real flights with four UAVs cut tension-tracking RMSE from 3.51 N to 2.09 N and still held agile figure-eight tracking.
arxiv.org/abs/2610.01185
6
Unknown rooms do not need a pre-built map.
LiDARFlow (ENAC): live LiDAR panels feed a lightweight guidance field so a MAV stays collision-free in clutter with only onboard sensing. Indoor flights finished waypoint and directional tasks in real time.
arxiv.org/abs/2610.01573
2
Landing-pad peak rate is not vertiport capacity.
UC Berkeley: takeoff/landing plus charging typically leave a single-TLOF site near half theoretical max when short queues must hold with high confidence. Extra pads and steadier turnarounds recover it.
arxiv.org/abs/2609.37288
4
KTH's battery-aware RL for aggressive quadrotor flight cut circle tracking error 49% at 3.84 m/s on a 38 g Crazyflie Brushless, and mean 20-lap race time from 106.22 s to 95.24 s. Per arXiv 2609.37316: the policy sees filtered pack voltage so it can use thrust that stock limits leave unused.
arxiv.org/abs/2609.37316
6
Zhejiang researchers train OmniRisk to score escape paths offline so a quadrotor can dodge onboard. Real flights held consecutive evasions at relative speeds up to 15 m/s with no fine-tuning (arXiv 2609.18191). For autonomous VTOL in clutter, reaction cost stays fixed as obstacle count rises.
arxiv.org/abs/2609.18191
4
Eindhoven researchers put a light pitot on a boom ahead of a multirotor so the controller sees gusts before they hit the airframe. Outdoor hover error along the wind fell 54% vs a PX4 baseline (arXiv 2609.31185). For autonomous VTOL, preview beats chasing the drift after the wind already moves you.
arxiv.org/abs/2609.31185
5
University of Dayton鈥檚 battery-aware planner cut multirotor energy by 7.46% on a 640 m, 150 s gust mission. Position error dropped about 72% vs a fixed wind-aware path. Gust fighting drains voltage, and low voltage starves the motors.
arxiv.org/abs/2609.12188
4
A CNN trained on real hexacopter flights found which rotor tip was damaged and how badly: over 99% rotor ID, about 96% severity within 卤1% (Oklahoma State, arXiv 2609.14180). For autonomous VTOL, that is the early warning layer before a tip nick becomes a tip-over.
arxiv.org/abs/2609.14180
3
















