POSE: Phase II: An Open-Source Ecosystem for Scenic

University Of California-Berkeley, Berkeley CA

Investigators

Abstract

The world is being transformed by the increasing use of autonomy, powered by artificial intelligence (AI) and machine learning (ML), across applications of great societal importance, including in transportation, energy, healthcare, and finance. Concurrently, there is increasing concern about the brittleness of AI and ML components and their susceptibility to failures that can compromise overall system safety. To address these concerns, the team of researchers have developed Scenic, an open-source language and toolkit supporting a systematic methodology for the design, verification, and deployment of AI systems. Scenic has been successfully demonstrated in multiple industrial-scale applications, including autonomous driving and avionics, and the open-source developer and user base is starting to grow. This project is developing a robust and sustainable open-source ecosystem (OSE) for Scenic. The project’s novelties include (i) developing a range of new applications and demonstrations of Scenic, (ii) developing a governance structure for the Scenic OSE, and (iii) creating sustainable Scenic OSE infrastructure. The project impacts the design of high-assurance AI systems by furthering the adoption of formal methods for the design of AI systems across industry, academia, and government. In addition to outreach to industry and government, the researchers are using Scenic and the Scenic OSE in educational outreach activities to undergraduate and high-school students. The project approach includes the following key components. First, the existing user base and collaborations in the domain of autonomous vehicles is being strengthened and extended. Second, end-user discovery is being performed to develop new high-impact applications of Scenic in virtual and augmented reality, healthcare, home and industrial robotics, multi-agent dynamic games, and other areas, along with corresponding networks of users and collaborators. Third, a governance structure for Scenic is being created comprising a steering committee, core team, and multiple working groups that engage with users, collaborators and stakeholders. Fourth, a sustainable Scenic OSE infrastructure is being developed by drawing on best practices for code quality and security, testing and verification, licensing, community interactions, maintenance and documentation, dissemination, etc. The team is also engaging in dissemination and outreach activities, conducting workshops, bootcamps, and tutorials, and producing documentation and other materials to support the Scenic OSE. Altogether, this NSF POSE project is working to establish Scenic as a strong open-source foundation enabling dependable societal-scale applications through high-assurance AI-enabled cyber-physical systems. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.