Organizer
Dr. Jonathan Petit, Senior Director of Research, OnBoard Security, Inc.
Session Description
The Automated Vehicles Ecosystem is composed of three components: the vehicle itself, an infrastructure, and users. The vehicle is composed of sensors, processing units, and uses machine learning algorithms and communication technologies. The infrastructure is made of network infrastructure (roadside unit, cellular base station), lane marking, traffic signs, traffic management center, and cloud services (to name a few entities). The users can be vehicle passengers, operators, remote operators, or vulnerable road users. Therefore, it is paramount to design an end-to-end security solution to ensure resilience of the whole system. Unfortunately, until now, we have seen some solutions (if any), but they are all designed in silos. In this session, experts of each subsystem will give you the state-of-the-art in security and privacy. We will discuss AV sensors security, machine learning security, human factors, road / fleet operator security. As policies / regulations play a major role in development and adoption of cybersecurity measures, we will also give an overview of the current legal framework. We will conclude the session with a panel discussion in order to put all solutions together and identify synergies.
Goals/Objectives/Outputs
Cybersecurity is commonly accepted as one of the three key enablers for AV. As the top conference in AV, AVS must have one breakout session on cybersecurity. This session will give a unique overview of the challenges faced by the whole ecosystem. By raising awareness of each subsystem issues, this session will help the cybersecurity community to work together, but will also influx cross-disciplinary research. With this session, we expect to identify common research needs between subsystems and help designing an end-to-end security solution.
Agenda
1:30 PM – 2:00 PM Semantic Adversarial Analysis and Learning with VerifAI
Adversarial analysis of deep neural networks has shown us that it is easy to find small
perturbations of inputs that produce undesired outputs. However, not all adversarial inputs
are meaningful. In this talk, we describe how formal methods can be used to define the
"semantic space" of inputs for a deep neural network used for perceptual tasks, to
analyze the network for desired semantic properties, and to re-train the network to achieve
improved performance. We present VerifAI, an open-source toolkit based on
this approach, showing how it can be used design and analyze autonomous vehicles
that use deep neural networks and other machine learning models.
- Shromona Ghosh, University of California, Berkeley
2:00 PM – 2:30 PM Security Considerations for ITS
With a shift toward leveraging emerging technologies to improve the safety and efficiency of the broader transportation system, attention must be given to potential security implications of deploying those technologies on a large scale. This talk will cover a sampling of security-related topics relevant to intelligent transportation systems (ITS), including connected and automated vehicles, as well as traffic management systems. Especially, I will detail our recent work on Over-The-Air update security and GPS security.
- Eric Thorn, Manager R&D, Cooperative Systems Section, Southwest Research Institute
2:30 PM – 3:00 PM Break
3:00 PM – 3:30 PM Safe and Secure Sensor Considerations in an Ever Evolving AV Threatscape
Sensor security offers new and unique considerations regarding investigating how sensor data may be manipulated adversarially or unintentionally to invoke undesired or unauthorized path planning and following behavior for one or many connected intelligent vehicles. In order to design more robust vehicle control systems in the future we must model these new consideration as part of a larger security framework and ecosystem. In this talk we explore the current vehicle security ecosystem and look to expand that ecosystem to include how scenery manipulation and other emerging vectors should be thought about, modelled and possibly mitigated as we continue to evolve.
- John Moore, Ford Motor Company, Advance Electrical, Software, Compute, and AI Vehicle Research Team
3:30 PM – 4:00 PM Using drivers to respond to vehicle cyberattacks
Increasing autonomy in vehicles is associated with many improvements in terms of functionality and convenience, but it also associated with potential cyberattacks as the attack surface of vehicles grows increasingly large. There isn’t a one-size-fits-all approach to prevent or mitigate vehicle cyberattacks. However, drivers are those who interact directly with the vehicle and have the ability to respond to unexpected situations caused by cyberattacks. I will describe results of an experimental study that examined how people responded to vehicle cyberattacks and I will present a prototype training and in-vehicle message systems that is designed to improve drivers’ awareness of and responses to vehicle cyberattacks. Using a ‘human in the loop’ approach introduces an innovative solution that addresses the human factor of vehicle cyberattacks.
- Shannon Roberts, Assistant Professor, University of Massachussets Amherst
4:00 PM – 4:30 PM The Unhackable AV That Totally Respects Privacy
Many ask "How can we create an unhackable AV that fully complies with all privacy requirements?" The answer is: we can’t. Our cybersecurity and privacy strategies must deal with the reality that there will be hacks and privacy breaches. But with the rapidly growing thicket of regulations and standards in the industry, it is not clear what policies and practices to develop to guard against and respond to those threats. This talk will present an overview of the current regulatory and legal frameworks for cybersecurity and privacy, describe emerging legal trends, and give practical tips for how to prepare for legal requirements and risks to AVs.
- Melody Drummond Hansen, Intellectual Property and Technology Partner, O’Melveny
4:30 PM – 5:30 PM Panel: End-to-End Security?
To seriously pretend to secure the automated vehicle ecosystem, we need to verify that all the solutions devised are interoperable. In this panel, speakers will analyze how the solutions fit together and potentially idenfity gaps that need to be addressed.
Moderator:
- Jonathan Petit, Senior Director of Research, OnBoard Security
Panelists:
- Melody Drummond Hansen, O’Melveny
- John Moore, Ford
- Shannon Roberts, Assistant Professor, University of Massachussets Amherst
- Eric Thorn, Southwest Research Institute
5:30 PM Conclusion
https://www.surveymonkey.com/r/C89WHB5