Neural Predictor for Flight Control with Payload
Ao Jin, Chenhao Li, Qinyi Wang, Ya Liu, Panfeng Huang, Fan Zhang
AI summary
Problem
Suspended payloads introduce unmodeled perturbations that degrade the closed-loop control performance of tethered UAVs. Existing methods struggle with computational complexity, lack interpretability, or require prior payload knowledge and dedicated force sensors.
Approach
The method uses a neural network to learn a lifted linear system from flight data, explicitly capturing external force and torque dynamics. This learned model is fused with nominal quadrotor dynamics and integrated into a Model Predictive Control framework for real-time flight control.
Key results
- Accurately models payload-induced forces/torques with bounded prediction error guarantees
- Reduces force and torque estimation errors by up to 66.15% and 33.33% versus state-of-the-art estimators
- Significantly improves closed-loop tracking performance in simulations and real-world flights
- Eliminates need for explicit payload modeling or force sensors while requiring fewer training samples
Why it matters
Enables reliable, high-precision aerial payload transport for logistics and inspection applications by removing the need for complex physical modeling and dedicated force sensors.
Abstract
Aerial robotics for transporting suspended payloads as the form of freely-floating manipulator are growinggreat interest in recent years. However, the force/torque caused by payload and residual dynamics will introduce unmodeled perturbations to the aerial robotics, which negatively affects the closed-loop perfor- mance. Different from estimation-like methods, this letter proposes Neural Predictor, a learning-based approach to model force/torque induced by payload and residual dynamics as a dynamical system. It yields a hybrid model that combines the first-principles dynamics with the learned dynamics. The hybrid model is then integrated into a MPC framework to improve closed-loop performance. Ef- fectiveness of proposed framework is verified extensively in both numerical simulations and real-world flight experiments. The re- sults indicate that our approach can capture force/torque caused by suspended payload and residual dynamics accurately, respond quickly to the changes of them and improve the closed-loop perfor- mance significantly. In particular, Neural Predictor outperforms a state-of-the-art learning-based estimator and has reduced the force and torque estimation errors by up to 66.15% and 33.33% while requiring less samples.