BRIDGES: Bridging the Semantic Gap between Safety-Critical Autonomous System Requirements and Semantically-Rich Sensor Data

  

Abstract: 

Autonomous Systems (ASs) carry out their safety-critical missions in complex environments. ASs sense their environment through complex, semantically-rich sensor inputs such as LiDAR point clouds and camera images. To safely complete its mission without casualties, the AS must have a sufficiently accurate and complete internal model of the world. Such models are created from sensor data through perception; for example, a semantic segmentation of an image infers the type of entity represented by each pixel. These models form the basis for higher-order reasoning, e.g., the rich body of research focused around leveraging such models for planning and control of the AS's behaviors. This dissertation tackles a different but important use of these models: verification and validation of the AS with respect to its safety requirements. These models provide a language over which to formally express and ultimately check AS safety requirements. The current challenge is that existing models are not fit for this purpose, leaving a significant gap between the languages available from these models and the languages necessary to express relevant safety requirements. In this work, I formalize this problem of aligning requirements and models over a common language that is within reach of modern AS perception. I introduce the parametric BRIDGES framework that structures the solution space, providing a framework to reason about the requirements that are expressible over different world models, and identifying critical gaps in the current state of the art. I further provide concrete instantiations of the framework targeting four classes of requirements to advance the state of the art in AS safety verification and validation.

Committee:  

  • Matthew Dwyer, Committee Chair, (CS/SEAS/UVA)
  • Sebastian Elbaum, Co-Advisor (CS/SEAS/UVA)
  • Kevin Sullivan, Co-Advisor (CS/SEAS/UVA)
  • Madhur Behl (CS, SYS/SEAS/UVA)
  • Nicola Bezzo (SIE, ECE/SEAS/UVA)
  • Marsha Chechik (University of Toronto Computer Science)