Knowledge-Aware Cyber-Physical Systems
Abstract:
During the development process of CPS, an analysis of whether the system operates safely in its target environment is of utmost importance. This entails two interconnected research goals in the research areas of system design and system verification, which tie together research in formal verification of CPS with research on knowledge representation and reasoning in multi-agent systems:
1. First of all, the perception of the world by agents and their internal representation of knowledge has to be captured. This should be explicit in the model in order to allow exploiting the distinction between the real world and the incomplete internal view of the CPS in reasoning about its properties.
2. In order to reason about the collective safety of multiple agents in a CPS without assuming worst- case behavior, knowledge about the goals, intents, and potential behaviors of interacting agents has to be captured. Such knowledge should be explicit in models in order to allow reasoning about the emerging behavior of multiple agents following their possibly conflicting goals.
These goals pose a challenging connection between epistemics used to model perception and multi-agent games as a mechanism to apply knowledge in a structured but a priori undetermined way. Communication and awareness are the keys to these interactions. Any zone of non-trivial traffic, be it sea, land or air, is likely to benefit from CPSs designed using this new epistemic hybrid game paradigm. For example:
1. Reduced congestion: by communicating sensed obstacles, cars and intentions, controllers can pack themselves more tightly, leading to more throughput without change in infrastructure.
2. Safer maneuvers: a car A decides when a safe lane switch can occur based on its own sensors and on the knowledge of car B in the lane that it is changing into, making B aware of A’s intentions. Thus, communication and mutual awareness determine safety and safety margins.
This project will extend existing logics with epistemic primitives for learning new information, and reasoning and making choices based on that information. It will give agents the ability to work towards achieving more flexible objectives that aren’t restricted to trying to satisfy or falsify a safety property.
Modeling knowledge explicitly is critical in bridging the gap between theoretical and practical safety of CPS. The perfect knowledge assumption of CPS controllers is largely impossible to implement in practice due to sensor noise and polling rates. With knowledge as a first-class citizen of CPS modeling languages it will be possible to realistically portray controller capabilities, and enable relevant and efficiency-increasing controllers for multi-agent cyber-physical systems.