Resilient Control of Cyber-Physical Systems with Distributed Learning - April 2022
PI(s) and Co-PI(s): Sayan Mitra and Geir Dullerud and Sanjay Shakkotai (U. Texas at Austin)
Researchers: Dawei Sun and Negin Musavi
HARD PROBLEM(S) ADDRESSED
This refers to Hard Problems, released November 2012.
Resiliency: Effective verification of safety and security properties of autonomous and cyber-physical systems
Metrics: How much data is necessary to achieve a certain level of confidence regarding a safety/security claim
PUBLICATIONS
Papers written as a result of your research from the current quarter only.
Papers in preparation and submission:
PAC Bounds for Generalization using Invariant Representations.
Advait Parulekar, Karthikeyan Shanmugam, Sanjay Shakkottai. Under submission, 2022.
Verifying Controllers with Convolutional Neural Network-based Perception: A Case for Intelligible, Safe, and Precise Abstractions. Chiao Hsieh, Keyur Joshi, Dawei Sun, Yangge Li, Sasa Misailovic, and Sayan Mitra. Under submission, 2022.
Asymptotically-Optimal Gaussian Bandits with Side Observations.
Alexia Atsidakou, Orestis Papadigenopoulos, Constantine Caramanis, Sanjay Shakkottai. Under submission, 2022.
NeuReach: Learning Reachability Functions from Simulations.
Dawei Sun and Sayan Mitra, To appear in the proceedings of Int. Conf. on Tools and Algorithms for Construction and Analysis of Systems (TACAS), 2022. Paper presented in April 2022.
Multi-agent Motion Planning from Signal Temporal Logic Specifications.
Dawei Sun, Jingkai Chen, Sayan Mitra, Chuchu Fan to appear in the proceedings of IEEE Robotics and Automation
Letters (RA-L), and Intl. Conf. on Robotics and Automation, May 2022.
Policy Optimization for Markovian Jump Linear Quadratic Control: Gradient-Based Methods and Global Convergence
and Parameter Synthesis for Stochastic Systems using Optimistic Optimization, Joao Jansch-Porto, Bin Hu, and Geir Dullerud, submitted for review, February 2022.
MLEFlow: Learning from His- tory to Improve Load Balancing in Tor, H. Darir, H. Sibai, C.-Y. Cheng, N. Borisov, G.E. Dullerud, and S. Mitra, to appear at Privacy Enhancing Technologies Symposium(PETS), 2022. Also presented at HoToS 2022.
Linear Bandit Algorithms with Sublinear Time Complexity, Shuo Yang, Tongzheng Ren, Sanjay Shakkottai, Eric Price, Inderjit Dhillon and Sujay Sanghavi, submitted for review, 2021.
Multi-Agent Low-Dimensional Linear Bandits, Ronshee Chawla, Abishek Sankararaman, and Sanjay Shakkottai. Submitted for review, 2021.
A Model-free Adversarial Reinforcement Learning Approach for mu Synthesis, by Darioush Keivan, Aaron Havens, Peter Seiler, Geir E. Dullerud, Bin Hu, accepted to appear in Proceedings of American Control Conference(ACC), 2022.
Revisiting PGD Attack for Stability Analysis of Large-Scale Nonlinear Systems and Perception-Based Control, by Aaron Havens, Darioush Keivan, Peter Seiler, Geir E. Dullerud, Bin Hu, Submitted for Review, 2022.
Low-fidelity Gradient Updates for High-fidelity Reprogrammable Iterative Learning Control, by Kuan-Yu Tseng, Jeff S. Shamma, Geir E. Dullerud, accepted to appear in Proceedings of American Control Conference(ACC), 2022.
KEY HIGHLIGHTS
Each effort should submit one or two specific highlights. Each item should include a paragraph or two along with a citation if available. Write as if for the general reader of IEEE S&P.
The purpose of the highlights is to give our immediate sponsors a body of evidence that the funding they are providing (in the framework of the SoS lablet model) is delivering results that "more than justify" the investment they are making.
We are developing safety and security analysis approaches for real-life of autonomous and cyber-physical systems using statistical and machine learning techniques. Our approaches rely on distributed and sample-efficient optimization techniques that have been developed in the context of the Multi-armed bandit problem. We have shown how these optimization algorithms can be used effectively for statistical model checking of markov decision processes and hybrid systems. We have built a suite of benchmarks related to online safety analysis of autonomous and semi-autonomous vehicles. Our initial results are very promising as the data usage and the running time of our algorithms can be several orders of magnitude better than existing model checking approaches such as Storm and Prism. Two PhD students are dedicating their research time to the project and the prototype tool has been made available online.
COMMUNITY ENGAGEMENTS
We developed and organized the GRAIC Autonomous Racing Competition which was co-located with CPSWeek 2021. The live event had more than 80 registered members and 30+ attendees. The software framework has been made available to the community for research.
We have been presenting monthly project updates to our project's champions Stephanie Polczynski and Jason Hogue.
Mitra
- Mitra served as the General Chair of HoTSoS 22. April 5-7 2022.
- Shakkottai has conducted a week-long workshop on Causal Inference, January 2022. This was a bootcamp that focused on models for causality, and their applications to machine learning.
- Shakkottai co-organized a workshop on Machine Learning for Systems, April 2022. This workshop hosted researchers from academia and industryto focus on technical challenges stemming from the deployment of ML pipelines at scale.
- Dullerud presentation at POSTECH, Korea; "Learning for Safety and Control in Dynamical Systems", April, 2022.
- A mini-version of the GRAIC Autonomous racing competition will be demonstrated to the public at the Engineering Open House at Illinois, May 8-9th 2022.
EDUCATIONAL ADVANCES
None to report.