Autonomous Vehicle Safety Technical and Social Issues
More than three decades of ground vehicle autonomy development have led to the current high profile deployment of autonomous vehicle technology on public roads. Ensuring the safety of these vehicles requires solving a number of challenging technical, social, and political problems. Significant progress can be made on control and planning safety via the use of a doer/checker architecture. Perception validation is more challenging, and thus far has primarily relied upon road testing for most developers. Even if closed course testing and simulation are increased, the problem of edge cases not seen in on-road data collection will remain due to the likelihood of a heavy tail distribution of surprises. Part of this heavy tail is subtle environmental degradation, which our work has shown can cause failures that reveal potential weak spots in perception systems. The talk will summarize my experiences in these areas as well as lay out the basis for the broader hard questions of how safe is safe enough, whether deployment delay cost lives, and the topic of regulation.
Prof. Philip Koopman
Carnegie Mellon University / Edge Case Research
Prof. Philip Koopman is a faculty member at the Carnegie Mellon University ECE department, with additional affiliations with the Institute for Software Research and the Robotics Institute. He leads research on safe and secure embedded systems and teaches cost-effective embedded system design techniques.
He has over 20 years of experience with autonomous vehicle safety, dating back to the CMU Navlab team and the Automated Highway Systems (AHS) program. His most recent projects include using stress testing and run time monitoring to ensure safety for a variety of vehicle and robotic applications for the research, industry, and defense sectors. He has additional experience with automotive and industrial functional safety, including testifying as an expert in vehicle safety class action litigation and consulting to NHTSA.
He is co-founder of Edge Case Research, which provides tools and services for autonomous vehicle testing and safety validation. His pre-university career includes experience as a US Navy submarine officer, embedded CPU designer at Harris Semiconductor, and embedded system architect at United Technologies. He is a Senior Member of IEEE, a Senior Member of ACM, and a member of SAE.
Challenges in the Qualification of Safety-Critical Machine Learning-based Components
The explosion of Machine Learning (ML) performance push strongly to consider their use in safety-critical embedded systems, such as autonomous vehicles. The assurance process of safety-critical software and systems in regulated industries like aerospace, nuclear power, railway or automotive is well streamlined and mastered for a long time already. These industries use well-defined standards, regulatory frameworks and processes, as well as formal techniques to assess and demonstrate the safety of the developed systems and software. However, the uncertainties and opaqueness of ML-based systems and components are difficult to validate and verify against most of traditional safety engineering methods. This raises the question of the complexity for integrating those components in safety-critical systems. This talk will explore the challenges and barriers for the integration of ML-based components in safety-critical systems. It will also provide insights about the main concerns reported by various collaboration cases with industry.
Prof. François Terrier
Commissariat à l’Energie Atomique (CEA)
François Terrier is head of the software and system engineering department at CEA LIST Institute. François Terrier holds a PhD in artificial intelligence and worked 10 years in the domain of expert systems using three-valued, temporal or fuzzy logics. Since 1994, he conducts research on software engineering. He was CEA representative in the European Network of Excellence on embedded systems and for OMG’s standardization. He conducts and leads research on model-driven engineering solutions for trustable systems and software.
CEA LIST, is a major hub for trustworthy digital systems and artificial intelligence, including safety and cybersecurity research. They develop tools to boost digital trust and make communications more secure, software to ensure aircraft operating security with an aeronautics-industry leader, formal methods to ensure software security with a major security-industry player. In Artificial Intelligence, CEA LIST, works with a major automotive manufacturer and its suppliers to develop embedded intelligence for autonomous vehicles.
The system and software engineering department’s activity is centered on the definition of methods and the development of tools for trustable systems. As head of the system and software engineering department, François Terrier is in charge of the new research program on trustworthy artificial intelligence for CEA LIST.