A Cross-Layer Framework for Retrofitting Robotic Vehicle Controllers
Abstract
Short work statementThis project proposes to develop a cross-layer framework to retrofit UxV controllers for high fault and attack resiliency. It targets legacy UxVs with binary-only control programs and can be adapted for other Naval ICCS~s. To retrofit a legacy system, the framework will enable four cross-layer capabilities, each to be realized by a research task that straddles the cyber and control domains:~Control Model Reverse Engineering (Task I) aims at inferring the control model (control algorithm + vehicle dynamics) of a subject system, based on control program reverse engineering and physical system identification.~Vulnerability Discovery (Task II) goes beyond traditional program vulnerabilities by identifying vulnerabilities in control models and hidden malicious logic in control programs.~Controller Retrofitting (Task III) involves retrofitting the control model and program, by enhancing the control algorithm, adding a shadow reference system for attack detection and recovery, and rewriting the binary control program without source code.~Assured Real-Time Operation (Task IV) is supported by timing analysis on both the control program and control model to preserve end-to-end real-time controllability.ObjectiveThe objective of the proposed project is to develop a cross-layer framework to retrofit UxV controllers for high fault and attack resiliency. It targets legacy UxVs with binary-only control programs and can be adapted for other Naval ICCS~s.ApproachThis project is to develop a framework to retrofit UxV controllers for high resiliency against faults and attacks. Different from traditional fault-tolerant computing and cybersecurity efforts, The PIs will take a cross-layer approach, based on the ~cyber-physical~ abstraction of a UxV. Existing approaches to ICCS security fall into two main categories: the cyber-centric approaches and the control-centric approaches. Solutions in the cyber-centric category tend to focus on the programs that implement the control, sensing, and actuation functions. While these cyber-centric approaches substantially advance the state-of-the-art, many of them do not involve in-depth analysis of a system~s control model, hence missing the opportunities to eliminate its control vulnerabilities and to leverage its physical properties. As a result, a system free of program vulnerabilities may still be exploitable via its control vulnerabilities. Solutions in the control-centric category tend to focus on the control models and methods. Robust control deals with uncertainties in the physical system and external disturbances. It involves quantification of uncertainty bounds and development of robust control algorithms under such bounded uncertainties. Adaptive control tackles uncertain and time-varying parameters of a system. It involves the design of adaptive controllers with varying parameters to accommodate real-time changes in plant parameters. Unfortunately, control-based approaches target physical failures or disturbances, not cyber-attacks. Most control-centric efforts do not address the emergence of cyber-physical attacks, which are launched via a cyber-attack-vector (e.g., malicious program logic) but inflict physical damages by manipulating the control model, parameters, and outputs. The proposed framework fosters inter-play between cyber security and resilient control research. The proposed efforts will produce concepts, methodologies, and software/hardware artifacts to instantiate a new paradigm: control-guided cyber-physical security. The novelty of the framework includes with the following salient features: (1) co-analysis in both cyber and control domains; (2) discovery of control model vulnerabilities and hidden malicious control program logic; (3) retrofitting both control model and program to defend against cyber-physical attacks. Overall merits & ONR mission relevanceCyber-physical-systems retrofitted with the tools that will be developed within this effort are expected to p
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jan 04, 2017
- Source ID
- N000141712045
Entities
People
- Dongyan Xu
Organizations
- Office of Naval Research
- United States Navy
- University of Virginia