Multi-Sensory Event Detection for Cross-Platform Coordination and Verification

With electronic devices increasingly interacting with each other and with the physical world around them, device behavior is becoming more and more like human behavior, namely reliant on experiences and not just data. Hence, there's a disconnect between how most devices have been engineered and what what practitioners are expecting them to do, and because of this fact, there's also a disconnect between the common cyber-security protections that protect data and some of the desirable proofs and verifications that can be highly valuable to cyber-physical devices. Our project is thus interested in studying the physical-world equivalents of certain cryptographic protections available in the cyber domain, namely the ability to provide a proof about a physical property to another device.

Toward this goal, we take a page out of the human behavior play-book, translating ideas from human perception to the cyber-physical domain, to provide stronger verification properties that will provide significant value for a variety of cyber-physical systems, devices, and scenarios. In general, proving a physical property is a hard problem, and no cryptographic primitive can address this problem itself (e.g., a device can sign telemetry data, but this only proves the source of the measurement, not its correctness). Several recent approaches rely on a large number of observers to collectively validate a measurement using majority voting, but most real-world scenarios do not support a large number of observers. We are not setting out to solve the broad and general problem of proving a physical property. Instead, we aim to carve out a smaller and more feasible sub-problem, namely how to use sensor measurements to provide contextual cues that allow us to verify certain physical aspects of a system. More specifically, given a physical property, we aim to identify contextual information that can provide valuable verification mechanisms for specific applications of interest. In our work, we consider applications in (1) home, commercial, and industrial Internet of Things (IoT) scenarios; (2) multi-vehicle coordination and cooperative driving scenarios; and (3) management aspects of smart cities (e.g., identification of communities to facilitate federation).

Through a combination of context-based event fingerprinting, device pairing, and verification, we aim to demonstrate the value of contextual proofs in select cyber-physical system domains that involve either physically-oriented group membership or other physical-world relationships. To this end, the goals of this project are: (1) to develop techniques to generate and verify "fingerprints" of the contextual properties of a set of devices, independent of hardware, vendor, and sensor types; (2) to realize mechanisms using context fingerprints to bootstrap trust in a secure and usable way among IoT devices in a variety of settings; and (3) to leverage fingerprinting primitives to enable verification of physical arrangement properties among coalitions of (semi-)autonomous vehicles. Beyond these initial goals, we have introduced tasks related to leveraging the same types of available data used in fingerprint generation (4) to characterize the source of the activity data (e.g., to identify the person performing the activities that are recorded) and (5) to provide value-added capabilities to existing contextual measurement-driven applications.