NNLander-VeriF: A Neural Network Formal Verification Framework for Vision-Based Autonomous Aircraft Landing

被引:13
作者
Santa Cruz, Ulices [1 ]
Shoukry, Yasser [1 ]
机构
[1] Univ Calif Irvine, Irvine, CA 92717 USA
来源
NASA FORMAL METHODS (NFM 2022) | 2022年 / 13260卷
基金
美国国家科学基金会;
关键词
Neural network; Formal verification; Perception;
D O I
10.1007/978-3-031-06773-0_11
中图分类号
V [航空、航天];
学科分类号
08 ; 0825 ;
摘要
In this paper, we consider the problem of formally verifying a Neural Network (NN) based autonomous landing system. In such a system, a NN controller processes images from a camera to guide the aircraft while approaching the runway. A central challenge for the safety and liveness verification of vision-based closed-loop systems is the lack of mathematical models that captures the relation between the system states (e.g., position of the aircraft) and the images processed by the vision-based NN controller. Another challenge is the limited abilities of state-of-the-art NN model checkers. Such model checkers can reason only about simple input-output robustness properties of neural networks. This limitation creates a gap between the NN model checker abilities and the need to verify a closed-loop system while considering the aircraft dynamics, the perception components, and the NN controller. To this end, this paper presents NNLander-VeriF, a framework to verify vision-based NN controllers used for autonomous landing. NNLander-VeriF addresses the challenges above by exploiting geometric models of perspective cameras to obtain a mathematical model that captures the relation between the aircraft states and the inputs to the NN controller. By converting this model into a NN (with manually assigned weights) and composing it with the NN controller, one can capture the relation between aircraft states and control actions using one augmented NN. Such an augmented NN model leads to a natural encoding of the closed-loop verification into several NN robustness queries, which state-of-the-art NN model checkers can handle. Finally, we evaluate our framework to formally verify the properties of a trained NN and we show its efficiency.
引用
收藏
页码:213 / 230
页数:18
相关论文
共 28 条
  • [1] [Anonymous], 2003, Introduction to Algorithms
  • [2] [Anonymous], 2012, An invitation to 3-d vision: from images to geometric models
  • [3] [Anonymous], Int. Verification of Neural Networks Competition (VNN
  • [4] Improved Geometric Path Enumeration for Verifying ReLU Neural Networks
    Bak, Stanley
    Hoang-Dung Tran
    Hobbs, Kerianne
    Johnson, Taylor T.
    [J]. COMPUTER AIDED VERIFICATION (CAV 2020), PT I, 2020, 12224 : 66 - 96
  • [5] Beard R.W., 2012, SMALL UNMANNED AIRCR, DOI DOI 10.1515/9781400840601
  • [6] Berre D.L., 2010, COMPUTER, V7, P59
  • [7] Clarke Edmund M, 2018, Handbook of Model Checking, P1, DOI [DOI 10.1007/978-3-319-10575-8_1, DOI 10.1007/978-3-319-10575-81, 10.1007/978-3-319-10575-8]
  • [8] Cruz US, 2021, Arxiv, DOI [arXiv:2104.02788, 10.48550/arXiv.2104.02788]
  • [9] Formal Verification of Piece-Wise Linear Feed-Forward Neural Networks
    Ehlers, Ruediger
    [J]. AUTOMATED TECHNOLOGY FOR VERIFICATION AND ANALYSIS (ATVA 2017), 2017, 10482 : 269 - 286
  • [10] Faugeras O., 1993, Three-dimensional Computer Vision: A Geometric Viewpoint