Autonomy and Counter-Autonomy
Evidence of autonomous system trustworthiness and resilience
Use of autonomy is providing leap-ahead capabilities as examples like intelligent parking assist and advanced emergency braking in modern automobiles show. For designers building autonomous features into software-based systems, issues of trust and emerging vulnerabilities loom large. The scarcity of techniques needed to assure software for autonomy presents a significant barrier to the broad, effective adoption and use of these advanced capabilities.
We perform and apply research to improve the development and effective use of partially or fully autonomous systems. To help humans understand why a system acts in a certain manner, we develop and train algorithms for explainable artificial intelligence. We are creating analysis and runtime monitoring approaches to reveal new types of vulnerabilities manifested in autonomy-enabled systems. And we are building the capability to ensure reliable operation in the face of malicious, counter-autonomous attack and manipulation by adversaries.
We created multi-agent planning techniques, middleware, and algorithms that enable single users to manage fleets of UASs in real-world environments with changing adversaries.
Distributed, adaptive real-time (DART) systems must satisfy safety-critical requirements. We developed a method to verify DART systems and generate assured code.
July 09, 2016 • Conference Paper
Stephanie RosenthalManuela Veloso (Carnegie Mellon University)Sai P. Selvaraj (Carnegie Mellon University)
In this work, we address the generation of narrations of autonomous mobile robot navigation experiences.Download
November 18, 2015 • Presentation
This presentation describes an evidence-based approach for producing high-assurance DART software involving multiple layers of the CPS stack.Download
August 19, 2014 • Presentation
This presentation summarizes the challenges surrounding Group Autonomy for Mobile Systems and how SEI research is addressing them.Download
July 08, 2013 • Conference Paper
In this paper, the authors advocate, formalize, and empirically justify an approach to compute quantitative utility of robotic missions using probabilistic model checking.Download