Verification of L1 Adaptive Control using Verse Library: A Case Study of Quadrotors

Lin Song, Yangge Li, Sheng Cheng, Pan Zhao, Sayan Mitra, Naira Hovakimyan

TL;DR

This paper explores the formal verification of L1 adaptive controllers (L1AC) for quadrotors using the Verse reachability analysis tool. They demonstrate that L1AC can achieve verifiable robust transient performance and gracefully handle input delays, a crucial step towards ensuring safety in autonomous systems.

What Problem Does This Paper Address?

Advanced Air Mobility (AAM) systems, like automated aircraft, demand high reliability. While L1 adaptive control (L1AC) is a promising technique for handling system uncertainties and guaranteeing performance, its theoretical guarantees haven't been formally verified for complex systems like quadrotors. This paper addresses the gap of formally verifying the transient performance and robustness of L1AC, specifically for an 18-dimensional quadrotor model, using a state-of-the-art verification tool.

Methods

The authors employ the Verse library, a Python tool for modeling and reachability analysis of hybrid, multi-agent systems. Verse uses a probabilistic algorithm (DryVR) for sensitivity functions and a simulation-based algorithm to over-approximate reachable states. The core method involves:

  1. Modeling: Representing the quadrotor's dynamics and the L1AC controller within the Verse framework. The quadrotor dynamics are described by standard equations of motion, and the L1AC is augmented to a geometric controller.
  2. Uncertainty Injection: Introducing model mismatch (e.g., different mass used for controller design vs. actual vehicle mass) and time delays in control inputs to simulate real-world conditions and test robustness.
  3. Reachability Analysis: Using Verse to compute the set of reachable states for the closed-loop system under these uncertainties and delays. This allows for formal verification of performance bounds.

Datasets

Not discussed in the paper. The authors use a simulated 18-dimensional quadrotor system with injected uncertainties and time delays as their "dataset" for evaluation.

Key Insights and Ideas

  • Formal Verification of L1AC is Achievable: The paper demonstrates that theoretical guarantees of L1AC, particularly its transient performance and robustness, can be formally verified using tools like Verse. This moves beyond simulation-based validation.
  • L1AC Handles Time-Varying Uncertainties Well: The verification shows that L1AC can maintain consistent and improved tracking performance even when faced with rapidly changing, uncertain system parameters (like mass). This highlights its adaptive capabilities.
  • Graceful Degradation with Input Delays: The study reveals that L1AC exhibits a predictable and graceful degradation of tracking performance as input delays increase, providing a quantifiable understanding of its robustness margins. This is crucial for designing systems that can tolerate communication or actuation latencies.

Key Results

  • Verifiable Robust Transient Performance: The L1AC system for the quadrotor was shown to achieve verifiable robust transient performance.
  • Effective Handling of Time-Varying Uncertainties: L1AC demonstrated consistent and better tracking performance compared to a baseline geometric controller when dealing with time-varying system parameters.
  • Input Delay Robustness: L1AC maintained reasonably good performance with input delays up to 60ms. The tracking performance degraded gracefully as the delay increased beyond this point.

Limitations and Future Work

  • Work-in-Progress: The authors explicitly state this is a work-in-progress abstract.
  • Future Work: The paper suggests extending the verification to systems incorporating learning-enabled components within the L1 adaptive controllers. They also aim to further explore the robustness margins and performance guarantees under more complex scenarios.

Summary updated

Abstract

L_1$ adaptive control ($L_1$AC) is a control design technique that can handle a broad class of system uncertainties and provide transient performance guarantees. In this work-in-progress abstract, we discuss how existing formal verification tools can be applied to check the performance of $L_1$AC systems. We show that the theoretical transient performance and robustness guarantees of an $L_1$ adaptive controller for an 18-dimensional quadrotor system can be verified using the recently developed Verse reachability analysis tool. We will further consider the performance verification of $L_1$AC on systems with learning-enabled components.
BibTeX
@inproceedings{song2023verification,
  author = {Lin Song and Yangge Li and Sheng Cheng and Pan Zhao and Sayan Mitra and Naira Hovakimyan},
  title = {{Verification of L1 Adaptive Control using Verse Library: A Case Study of Quadrotors}},
  booktitle = {ICCPS},
  year = {2023},
  doi = {10.1145/3576841.3589618},
}
Figure