Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

RoVerFly: Robust and Versatile Implicit Hybrid Control of Quadrotor-Payload Systems

Mintae Kim, Jiaze Cai, Koushil Sreenath

TL;DR

RoVerFly is presented, a unified learning-based control framework where a single reinforcement learning (RL) policy functions as an implicit hybrid controller, managing complex dynamics without explicit mode detection or controller switching.

Abstract

Designing robust controllers for precise trajectory tracking with quadrotors is challenging due to nonlinear dynamics and underactuation, and becomes harder with flexible cable-suspended payloads that add degrees of freedom and hybrid dynamics. Classical model-based methods offer stability guarantees but require extensive tuning and often fail to adapt when the configuration changes-when a payload is added or removed, or when its mass or cable length varies. We present RoVerFly, a unified learning-based control framework where a single reinforcement learning (RL) policy functions as an implicit hybrid controller, managing complex dynamics without explicit mode detection or controller switching. Trained with task and domain randomization, the controller is resilient to disturbances and varying dynamics. It achieves strong zero-shot generalization across payload settings-including no payload as well as varying mass and cable length-without re-tuning, while retaining the interpretability and structure of a feedback tracking controller. Code and supplementary materials are available at https://github.com/mintaeshkim/roverfly.

RoVerFly: Robust and Versatile Implicit Hybrid Control of Quadrotor-Payload Systems

TL;DR

RoVerFly is presented, a unified learning-based control framework where a single reinforcement learning (RL) policy functions as an implicit hybrid controller, managing complex dynamics without explicit mode detection or controller switching.

Abstract

Designing robust controllers for precise trajectory tracking with quadrotors is challenging due to nonlinear dynamics and underactuation, and becomes harder with flexible cable-suspended payloads that add degrees of freedom and hybrid dynamics. Classical model-based methods offer stability guarantees but require extensive tuning and often fail to adapt when the configuration changes-when a payload is added or removed, or when its mass or cable length varies. We present RoVerFly, a unified learning-based control framework where a single reinforcement learning (RL) policy functions as an implicit hybrid controller, managing complex dynamics without explicit mode detection or controller switching. Trained with task and domain randomization, the controller is resilient to disturbances and varying dynamics. It achieves strong zero-shot generalization across payload settings-including no payload as well as varying mass and cable length-without re-tuning, while retaining the interpretability and structure of a feedback tracking controller. Code and supplementary materials are available at https://github.com/mintaeshkim/roverfly.

Paper Structure

This paper contains 33 sections, 26 equations, 5 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Model-Based Control and Trajectory Planning for Quadrotor-Payload Systems
  4. Learning-based Control of Quadrotors
  5. Problem Description
  6. System Dynamics
  7. Taut Mode Dynamics ($T>0$)
  8. Slack Mode Dynamics and Switching Conditions
  9. Arbitrary Trajectory Tracking
  10. End-to-End Model-Free RL for Quadrotor Payload Trajectory Tracking
  11. Preliminaries
  12. Algorithm and Training Overview
  13. Belief-Based Analysis of I/O History for Implicit Hybrid Control
  14. Belief-Based Policy and Value under Uncertainty
  15. Hybrid Dynamics and Implicit Mode Estimation
  16. ...and 18 more sections

Figures (5)

  • Figure 1: Diagram of the training loop. Present features are concatenated with history/preview embeddings and fed to $\pi$. Collective thrust and body rates (CTBR) commands pass through the rate loop and mixer before actuation.
  • Figure 2: Zero-shot performance across payload mass $m_P$ and cable length $l$ on unseen trajectories.
  • Figure 3: Payload tracking on unseen trajectories: without payload and with flexible cable-suspended payload.
  • Figure 4: Hover initial perturbation rejection: position error traces ($l{=}1.0~\mathrm{m},~m_P{=}0.2~\mathrm{kg}$).
  • Figure 5: Controller behavior during slack-to-taut transition. (a) The controller modulates thrust before tautening, mitigating impact. (b) The payload exhibits distinct free-fall and bounce slack phases before settling. The shaded intervals highlight free-fall and bounce phases during slack-to-taut transition.