A carbon fiber pole stood on a tip the size of a coin for more than three hours in a windy Munich park, held up by a self-balancing pole drone. Four drone motors kept it upright, and they barely had to work. Aerospace engineer Peter Ryseck built the pole to test a control problem that is harder than it looks.
Ryseck is a flight dynamics engineer at Joby Aviation, and he holds a master’s degree from the University of Maryland. He started out building RC planes in his parents’ basement, a start that led him into drone engineering. The self-balancing pole drone project, he said, showed him how much control theory hides behind a simple-looking stick: “Balancing a pole turned out to be a lot harder than I was expecting.”
Why the first self-balancing pole drone design failed
The pole’s weight travels straight down the shaft into the ground. The motors on top only push sideways, in small corrections, whenever the pole begins to tip. Most of the time they run just above idle.
Ryseck’s first version used large, slow-turning 10-inch propellers, chosen for efficiency and low noise. The motors produced enough thrust, but they could not change it fast enough. When the pole leaned, the controller called for a correction, and the blades lagged behind. The pole oscillated and then hung there.
Smaller 3-inch propellers lacked the force to recover a falling pole. Ryseck settled on 5-inch props, the size common in FPV racing. They were heavy enough to catch a fall and light enough to spin up before the pole passed the point of recovery.
If drone design is your interest, Flafa’s camera-free drone design for aerial photography takes a very different approach to the airframe.
Taming the flex in the pole
The controller is a proportional-derivative loop. It reads how far the pole has tilted and how fast it is tilting, then responds to both. Ryseck tuned the gains in a computer simulation based on the pole’s measured mass and inertia. He wanted one set of numbers that would handle a longer shaft, a heavier load, such as a phone for a selfie, and gusts of wind without retuning, so that the self-balancing pole drone would stay stable across different setups.
The simulation assumed a rigid shaft, and the real one bends. Flight logs showed a structural resonance near 18 hertz. The actual balancing motion sits below about 5 hertz. Without a fix, the controller mistook the bending for falling and kept feeding energy into the flex.
Ryseck added a low-pass filter, which smooths out fast signals. He kept it limited, because it adds lag near the balancing frequencies. He also added a dynamic notch filter. It measures the pole’s natural frequencies each time the system arms and removes them from the control signal. The motor tone shifts in the video as each notch engages.
Leaning into the wind
Holding the motors at zero throttle sounds efficient, but it caused small instabilities. Ryseck kept a minimum throttle so the propellers always spin. The flight software also adjusts its estimate of vertical over time, because the inertial sensor never measures perfectly. In steady wind, that same logic lets the pole lean a few degrees into the breeze. Gravity then does most of the work.
Ryseck ran the finished self-balancing pole drone in Munich’s Olympiapark on a windy day and walked away. It was still standing after three hours. The battery, a 2200 mAh three-cell pack, dropped from a full 12.6 volts to 11.1 volts. DroneXL noted that one secondhand write-up misread the video and reported a run of 3.75 hours. The 3.75 figure in the video is the per-cell voltage.
Troy Wigton contributed some of the control ideas. Push the pole, and the props respond with a soft correction. The shaft straightens and returns to doing almost nothing.
What the project leaves unanswered
The video does not give the pole’s length or total weight. It does not show the power the motors draw at idle, and it does not record how strong the wind was during the park test. Ryseck has not said what the pole is for. He asks viewers what they think of it.
Balance problems also appear in walking machines, and ETH Zurich’s five-fingered robotic hand that walks shows another way engineers wrestle with control and stability.
The idle power figure matters most. It determines whether this self-balancing pole drone runs for three hours or for days. Until Ryseck publishes that number, the pole remains a well-documented demonstration of control theory, and a strong one.
Frequently asked questions
Why did the first self-balancing pole drone design fail?
The first version used large 10-inch propellers that could not change thrust fast enough when the pole leaned, so the pole oscillated and then hung there. The smaller 3-inch propellers lacked the force to recover a falling pole.
What propeller size did the final self-balancing pole drone use?
The final design used 5-inch propellers, the size common in FPV racing. They were heavy enough to catch a fall and light enough to spin up before the pole passed the point of recovery.
How did the engineer stop the pole’s flex from destabilizing the controller?
Flight logs showed a structural resonance near 18 hertz, and the controller mistook the bending for falling. The engineer added a low-pass filter and a dynamic notch filter, which measures the pole’s natural frequencies at each arming and removes them from the control signal.
How long did the carbon fiber pole stay upright in the test?
The pole stayed upright for more than three hours in a windy park test in Munich’s Olympiapark. During that run, the 2200 mAh three-cell battery dropped from 12.6 volts to 11.1 volts.
More in
Niche
All Niche
