Thu 25 April 2019

Open Source Safety Requirements Analysis for Autonomous Vehicles based on STPA

by Paul Sherwood , 2019 , Tags Open Source STPA Safety Autonomous vehicle machine learning


As a result of collaborative research with MIT, Codethink announces a new open source project AV-STPA, to analyse and document safety requirements for autonomous vehicles by applying System-Theoretic Process Analysis.

On our continuing journey to understand the implications of safety and security risk management for complex software-intensive systems, Codethink has been inspired by the STAMP/STPA approach pioneered by Professor Nancy Leveson and her colleagues at MIT.

Following discussions with Nancy and other members of the safety community in 2018, Codethink has been collaborating with John Thomas, MIT Executive Director of System Safety and Cybersecurity, in applying the STPA process to analyse safety for an Autonomous Vehicle software platform.

A key innovation on this project is that all of Codethink's work has followed an open source approach, with documented evolution tracked over time using Git to record work-in-progress and history. The Git Repo is here. As an open source project the work is available for review and study by interested parties around the world. As far as we know this is the first public initiative to document detailed system-level safety requirements for review and re-use, free from non-disclosure restrictions and commercial paywalls.

John Thomas presented findings from this research collaboration at the recent STAMP Workshop at MIT in Boston, attended by around 400 safety and cybersecurity professionals. The work was very warmly received, and many attendees commented that they appreciated that this project was able to discuss the details of the analysis as a direct consequence of the open-source approach adopted by Codethink.

A key point is that the STPA approach is top-down, starting from the general case and progressing towards specific implementation details. As a result we believe that most of the analysis can be be considered as a generically applicable input for autonomous vehicle safety architecture implementation. We hope that experts will review the public materials and contribute where possible, to establish a useful representation of the 'state-of-the-art'.

The repository is a work-in-progress - incomplete and not perfect; however it represents a step forward in our general understanding of the safety requirements for autonomous vehicle systems, which we believe can be of benefit to engineers and executives working on initiatives in this challenging domain. As with all open source projects, we welcome reviews and feedback as well as contributions to improve the work itself. In addition we are ready to share knowledge and assist in open analyis of safety and security requirements for other domains.

The original upload of the article can be found here