Online Learning and Control for Data-Augmented Quadrotor Model

Matej Smid,Jindrich Dunik

2023-04-02

Abstract:The ability to adapt to changing conditions is a key feature of a successful autonomous system. In this work, we use the Recursive Gaussian Processes (RGP) for identification of the quadrotor air drag model online, without the need of training data. The identified drag model then augments a physics-based model of the quadrotor dynamics, which allows more accurate quadrotor state prediction with increased ability to adapt to changing conditions. This data-augmented physics-based model is utilized for precise quadrotor trajectory tracking using the suitably modified Model Predictive Control (MPC) algorithm. The proposed modelling and control approach is evaluated using the Gazebo simulator and it is shown that the proposed approach tracks a desired trajectory with a higher accuracy compared to the MPC with the non-augmented (purely physics-based) model.

Robotics,Systems and Control

What problem does this paper attempt to address?

The problem that this paper attempts to solve is in quadrotor control, how to improve the model's adaptability to environmental changes through online learning, especially for the factor of air drag which is difficult to model accurately. Specifically, the author proposes a method based on Recursive Gaussian Processes (RGP) for online identification of the air drag model of the quadrotor without pre - collecting training data. This method can enhance the physics - based model, thereby improving the state prediction accuracy of the quadrotor and increasing its adaptability to changing conditions. By combining this enhanced data - driven model with Model Predictive Control (MPC), the paper aims to achieve more accurate quadrotor trajectory tracking.

The main contributions of the paper include:
1. **Online learning of air drag model**: Use RGP to online identify the air drag model of the quadrotor without pre - collecting training data.
2. **Enhanced physical model**: Combine the identified air drag model with the physics - based model to form a data - augmented physics - based model.
3. **Improved trajectory tracking**: Utilize the enhanced physical model and MPC algorithm to achieve more accurate quadrotor trajectory tracking.

Through experiments in the Gazebo simulator, the paper verifies the effectiveness of the proposed method, showing that compared with MPC using only the physical model, this method can track the desired trajectory more accurately.