Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Aggressiveness-Aware Learning-based Control of Quadrotor UAVs with Safety Guarantees

Leonardo Colombo, Thomas Beckers, Juan Giribet

TL;DR

This paper presents an aggressiveness-aware control framework for quadrotor UAVs that integrates learning-based oracles to mitigate the effects of unknown disturbances and provides a principled way to exploit learning for safer and less aggressive quadrotor maneuvers.

Abstract

This paper presents an aggressiveness-aware control framework for quadrotor UAVs that integrates learning-based oracles to mitigate the effects of unknown disturbances. Starting from a nominal tracking controller on $\mathrm{SE}(3)$, unmodeled generalized forces and moments are estimated using a learning-based oracle and compensated in the control inputs. An aggressiveness-aware gain scheduling mechanism adapts the feedback gains based on probabilistic model-error bounds, enabling reduced feedback-induced aggressiveness while guaranteeing a prescribed practical exponential tracking performance. The proposed approach makes explicit the trade-off between model accuracy, robustness, and control aggressiveness, and provides a principled way to exploit learning for safer and less aggressive quadrotor maneuvers.

Aggressiveness-Aware Learning-based Control of Quadrotor UAVs with Safety Guarantees

TL;DR

This paper presents an aggressiveness-aware control framework for quadrotor UAVs that integrates learning-based oracles to mitigate the effects of unknown disturbances and provides a principled way to exploit learning for safer and less aggressive quadrotor maneuvers.

Abstract

This paper presents an aggressiveness-aware control framework for quadrotor UAVs that integrates learning-based oracles to mitigate the effects of unknown disturbances. Starting from a nominal tracking controller on , unmodeled generalized forces and moments are estimated using a learning-based oracle and compensated in the control inputs. An aggressiveness-aware gain scheduling mechanism adapts the feedback gains based on probabilistic model-error bounds, enabling reduced feedback-induced aggressiveness while guaranteeing a prescribed practical exponential tracking performance. The proposed approach makes explicit the trade-off between model accuracy, robustness, and control aggressiveness, and provides a principled way to exploit learning for safer and less aggressive quadrotor maneuvers.
Paper Structure (17 sections, 3 theorems, 58 equations, 4 figures, 1 table)

This paper contains 17 sections, 3 theorems, 58 equations, 4 figures, 1 table.

Table of Contents

  1. Introduction
  2. System Modeling and Nominal Control
  3. System class
  4. Motivating Example
  5. Problem Setting
  6. Aggressiveness-Aware Learning Control
  7. Oracle and Model Error Bounds
  8. Gaussian Process as Oracle
  9. Aggressiveness-Aware Stability Guarantees
  10. Implementation Procedure
  11. Numerical Results
  12. Simulation Setup
  13. Performance--Aggressiveness Trade-off
  14. Moderate disturbance ($\texttt{DIST\_SCALE}=1$).
  15. Severe disturbance ($\texttt{DIST\_SCALE}=3$).
  16. ...and 2 more sections

Key Result

Theorem 1

Consider the system for:se3affine together with the learning-based control law for:framework:ctrl. Suppose that Assumptions ass:nominal_lyap and ass:gradV hold, and that the oracle satisfies Assumption ass:errorbound on a compact set $\mathcal{X}_c\subset\mathcal{X}$ containing the closed-loop traje where $c_1,c_2$ correspond to the Lyapunov inequalities for the chosen $H_N$. Denote by $\mathcal{E

Figures (4)

  • Figure 1: Tracking performance under fixed-low, fixed-high, and aggressiveness-aware gain selection. The dashed line denotes the tolerance $\varepsilon=0.1~\mathrm{m}$.
  • Figure 2: Aggressiveness metrics under fixed-low, fixed-high, and aggressiveness-aware gain selection. The shaded region marks the transient window $[0,3]~\mathrm{s}$ used for transient RMS.
  • Figure 3: Numerical validation under severe mismatch ($\texttt{DIST\_SCALE}=3$): tracking performance for fixed-low, fixed-high, and GP-comp (aware). The dashed line indicates the tracking tolerance $\varepsilon=0.10$ m.
  • Figure 4: Online experiment, DIST_SCALE$=3.0$. Position tracking error $\|e_p(t)\|$ for fixed-low (no GP), offline GP compensation, and offline+online residual GP. The dashed line denotes the tolerance $\varepsilon=0.10$ m.

Theorems & Definitions (10)

  • Remark 1
  • Remark 2
  • Remark 3
  • Remark 4
  • Remark 5
  • Remark 6
  • Remark 7
  • Theorem 1
  • Corollary 1
  • Proposition 1