Reinforcement learning policies for quadrotor inversion transitions with bidirectional thrust outperform optimization baselines by 32% in position RMSE and 57% in settling time in simulation, with successful hardware validation.
Multicopter attitude control for recovery from large disturbances
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
abstract
We present a novel, high-performance attitude control law for multicopters, with a view to recovery from large disturbances. The controller is compared to three well-established alternatives from the literature. All controllers considered are identical to first order, but differ in their computation of the attitude error. We show that the popular use of the skew-symmetric part of the rotation matrix is problematic from a safety perspective, and specifically that the closed loop system may linger at large attitude errors for an arbitrary duration (leading to potential failures of practical systems). The novel proposed controller prioritizes the error in the vehicle thrust direction, and is shown to outperform a similar, existing controller from the literature. Stability follows via a Lyapunov function, and the controller is validated in experiments. This novel controller is especially attractive in safety-critical situations, where a multicopter may be required to recover from large initial disturbances.
fields
cs.RO 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
AcroRL: Learning Aggressive Quadrotor Inversion using Bidirectional Thrust
Reinforcement learning policies for quadrotor inversion transitions with bidirectional thrust outperform optimization baselines by 32% in position RMSE and 57% in settling time in simulation, with successful hardware validation.