New MPC+GA method balances flying inverted pendulum with unknown length
A genetic algorithm estimator pairs with MPC to stabilize an unpredictable pendulum in flight.
A new paper from researchers Esther Paul, Mitchell Torok, and Mohammad Deghat tackles the classic inverted pendulum control problem with a twist: the pendulum is flying and its length is unknown. The team pairs an online Genetic Algorithm (GA) estimator with Model Predictive Control (MPC), allowing the system to simultaneously estimate the pendulum's physical parameters and compute optimal control actions in real time. This combination matters because traditional MPC assumes a known, fixed plant model — a luxury that many real-world robotic and aerial systems don't have.
The researchers validated their approach on a practical flying inverted pendulum rig, not just in simulation. They ran disturbance rejection tests and trajectory tracking experiments to measure controller robustness and tracking accuracy. The online GA estimator was also tested from multiple initial conditions, confirming both convergence speed and estimation accuracy. The work has been accepted for presentation at the 23rd IFAC World Congress, a leading venue for control systems research, and is currently available on arXiv under ID 2608.17256.
- Online GA estimator jointly estimates unknown pendulum length while MPC computes real-time control actions.
- Validated on a physical flying inverted pendulum system, not just simulation, using disturbance and trajectory tests.
- Accepted at the 23rd IFAC World Congress; paper available at arXiv:2608.17256.
Why It Matters
Real-time adaptive control for drones and robotics, enabling stable operation without precise system models.