Safe and Reliable Training of Learning-Based Aerospace Controllers

“Safe and Reliable Training of Learning-Based Aerospace Controllers” by Udayan Mandal, Guy Amir, Haoze Wu, Ieva Daukantas, Fletcher Lee Newell, Umberto Ravaioli, Baoluo Meng, Michael Durling, Kerianne Hobbs, Milan Ganai, Tobey Shim, Guy Katz, and Clark Barrett. In 2024 AIAA DATC/IEEE 43^rd Digital Avionics Systems Conference (DASC '24), Sep. 2024. San Diego, CA.

Abstract

In recent years, deep reinforcement learning (DRL) approaches have generated highly successful controllers for a myriad of complex domains. However, the opaque nature of these models limits their applicability in aerospace systems and sasfety-critical domains, in which a single mistake can have dire consequences. In this paper, we present novel advancements in both the training and verification of DRL controllers, which can help ensure their safe behavior. We showcase a design-for-verification approach utilizing k-induction and demonstrate its use in verifying liveness properties. In addition, we also give a brief overview of neural Lyapunov Barrier certificates and summarize their capabilities on a case study. Finally, we describe several other novel reachability-based approaches which, despite failing to provide guarantees of interest, could be effective for verification of other DRL systems, and could be of further interest to the community.

Keywords: Training;Space vehicles;Neural networks;Aerospace electronics;Deep reinforcement learning;Control systems;Aerospace safety;Reliability;Reachability analysis;Artificial intelligence;AI Safety;Deep Reinforcement Learning;Formal Verification;Deep Neural Network Verification

BibTeX entry:

@inproceedings{MAW+DASC24,
   author = {Udayan Mandal and Guy Amir and Haoze Wu and Ieva Daukantas
	and Fletcher Lee Newell and Umberto Ravaioli and Baoluo Meng and
	Michael Durling and Kerianne Hobbs and Milan Ganai and Tobey Shim
	and Guy Katz and Clark Barrett},
   title = {Safe and Reliable Training of Learning-Based Aerospace
	Controllers},
   booktitle = {2024 AIAA DATC/IEEE {\it 43^{rd}} Digital Avionics Systems
	Conference (DASC '24)},
   publisher = {IEEE},
   month = sep,
   year = {2024},
   doi = {10.1109/DASC62030.2024.10749499},
   note = {San Diego, CA},
   url = {https://ieeexplore.ieee.org/document/10749499}
}

(This webpage was created with bibtex2web.)