Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Proto-MPC: An Encoder-Prototype-Decoder Approach for Quadrotor Control in Challenging Winds

Yuliang Gu, Sheng Cheng, Naira Hovakimyan

TL;DR

The paper tackles robust quadrotor control under dynamic wind disturbances by introducing Proto-MPC, a model predictive control framework augmented with an Encoder-Prototype-Decoder (EPD) that learns residual dynamics. The EPD employs a universal encoder to capture shared task structure and task-specific linear prototype decoders to adapt quickly to individual wind conditions, with an online mechanism to interpolate among prototypes. A prototype-based meta-update balances cross-task knowledge with task specificity, enabling fast online adaptation even without privileged task information. Experiments in simulation show Proto-MPC achieves superior trajectory tracking under static and spatially varying winds while reducing online computation relative to full online retraining baselines, indicating strong generalization and practical potential for real-world deployment.

Abstract

Quadrotors are increasingly used in the evolving field of aerial robotics for their agility and mechanical simplicity. However, inherent uncertainties, such as aerodynamic effects coupled with quadrotors' operation in dynamically changing environments, pose significant challenges for traditional, nominal model-based control designs. We propose a multi-task meta-learning method called Encoder-Prototype-Decoder (EPD), which has the advantage of effectively balancing shared and distinctive representations across diverse training tasks. Subsequently, we integrate the EPD model into a model predictive control problem (Proto-MPC) to enhance the quadrotor's ability to adapt and operate across a spectrum of dynamically changing tasks with an efficient online implementation. We validate the proposed method in simulations, which demonstrates Proto-MPC's robust performance in trajectory tracking of a quadrotor being subject to static and spatially varying side winds.

Proto-MPC: An Encoder-Prototype-Decoder Approach for Quadrotor Control in Challenging Winds

TL;DR

The paper tackles robust quadrotor control under dynamic wind disturbances by introducing Proto-MPC, a model predictive control framework augmented with an Encoder-Prototype-Decoder (EPD) that learns residual dynamics. The EPD employs a universal encoder to capture shared task structure and task-specific linear prototype decoders to adapt quickly to individual wind conditions, with an online mechanism to interpolate among prototypes. A prototype-based meta-update balances cross-task knowledge with task specificity, enabling fast online adaptation even without privileged task information. Experiments in simulation show Proto-MPC achieves superior trajectory tracking under static and spatially varying winds while reducing online computation relative to full online retraining baselines, indicating strong generalization and practical potential for real-world deployment.

Abstract

Quadrotors are increasingly used in the evolving field of aerial robotics for their agility and mechanical simplicity. However, inherent uncertainties, such as aerodynamic effects coupled with quadrotors' operation in dynamically changing environments, pose significant challenges for traditional, nominal model-based control designs. We propose a multi-task meta-learning method called Encoder-Prototype-Decoder (EPD), which has the advantage of effectively balancing shared and distinctive representations across diverse training tasks. Subsequently, we integrate the EPD model into a model predictive control problem (Proto-MPC) to enhance the quadrotor's ability to adapt and operate across a spectrum of dynamically changing tasks with an efficient online implementation. We validate the proposed method in simulations, which demonstrates Proto-MPC's robust performance in trajectory tracking of a quadrotor being subject to static and spatially varying side winds.
Paper Structure (17 sections, 9 equations, 9 figures, 2 tables, 2 algorithms)

This paper contains 17 sections, 9 equations, 9 figures, 2 tables, 2 algorithms.

Table of Contents

  1. Introduction
  2. Background: Nonlinear MPC for Quadrotor Control
  3. Method
  4. Dataset
  5. Prototype-Decoder-Based Meta-Learning
  6. Prototype Decoder
  7. Encoder Meta-Update based on Prototype Decoder
  8. Proto-MPC
  9. Experiments
  10. Conclusion
  11. Appendix
  12. Related Work
  13. Algorithms
  14. Experimental Setup
  15. Learning Results
  16. ...and 2 more sections

Figures (9)

  • Figure 1: Framework Overview. a) Collecting data on multiple tasks; b) Pretraining to ensure that encoder-decoder pairs can capture the overall patterns of the data; c) Jointly training task-specific prototype decoders to capture distinctive task features and regularizing the encoder to avoid overfitting; d) Online implementation of Proto-MPC with prototype-decoder-based adaptation.
  • Figure 2: Illustration of the statistical model of task distribution.
  • Figure 3: Illustration of meta update. Blue indicates the weighted gradients and red indicates the update direction
  • Figure 4: Spatially varying wind distribution.
  • Figure 5: Tracking performance subject to spatially varying winds on different trajectories. The first row (\ref{['fig:path 1 mpc']}, \ref{['fig:path 2 mpc']}, \ref{['fig:path 3 mpc']}) shows the tracking performance of MPC with the nominal model, the second row (\ref{['fig:path 1 knode']}, \ref{['fig:path 2 knode']}, \ref{['fig:path 3 knode']}) shows the tracking performance of KNODE-MPC-Online (with spectral normalization) and the third row (\ref{['fig:path 1 proto']}, \ref{['fig:path 2 proto']}, \ref{['fig:path 3 proto']}) shows the tracking performance of Proto-MPC. The colorbar highlights the deviation from the reference trajectory
  • ...and 4 more figures

Theorems & Definitions (3)

  • definition 1
  • definition 2
  • remark 1