Autonomous Vehicle Safety: An Interdisciplinary Challenge

IEEE Intelligent Transportation Systems Magazine ( Volume: 9, Issue: 1, Spring 2017, Pp. 90-96).

Abstract: Ensuring the safety of fully autonomous vehicles requires a multi-disciplinary approach across all the levels of functional hierarchy, from hardware fault tolerance, to resilient machine learning, to cooperating with humans driving conventional vehicles, to validating systems for operation in highly unstructured environments, to appropriate regulatory approaches. Significant open technical challenges include validating inductive learning in the face of novel environmental inputs and achieving the very high levels of dependability required for full-scale fleet deployment. However, the biggest challenge may be in creating an end-to-end design and deployment process that integrates the safety concerns of a myriad of technical specialties into a unified approach.

Introduction:

A typical prediction of the future of autonomous vehicles includes people being relieved from the stress of daily commute driving, perhaps even taking a nap on the way to work. This is expected to be accompanied by a dramatic reduction in driving fatalities due to replacing imperfect human drivers with (presumably) better computerized autopilots. City governments apparently believe this will happen within 10 years (Boston Consulting Group 2015). But, how to get such fully autonomous vehicles to actually be safe is no simple matter (Luettel 2012, Gomes 2014). We outline a number of areas which present significant challenges to creating acceptably safe, fully autonomous vehicles compared to the vehicles of even a few years ago, with an emphasis on the difficulty of validating autonomy at the scale of a full-size vehicle fleet.

The question is not whether autonomous vehicles will be perfect (they won’t). The question is when we be able to deploy a fleet of fully autonomous driving systems that are actually safe enough to leave humans completely out of the driving loop. The challenges are significant, and span a range of technical and social issues for both acceptance and deployment (Rupp 2010, Bengler 2014, Learner 2015). A holistic solution will be needed, and must of necessity include a broad appreciation for the range of challenges (and potential solutions) by all the relevant stakeholders and disciplines involved.

Our work in building safety arguments and run-time safety mechanisms for autonomous ground vehicles has taught us that even understanding what “safe” really means for autonomous vehicles is not so simple. “Safe” means at least correctly implementing vehicle-level behaviors such as obeying traffic laws (which can vary depending upon location) and dealing with non-routine road hazards such as downed power lines and flooding. But it also means things such as fail-over mission planning, finding a way to validate inductive-based learning strategies, providing resilience in the face of likely gaps in early-deployed system requirements, and having an appropriate safety certification strategy to demonstrate that a sufficient level of safety has actually been achieved.

Read the rest here: https://ieeexplore.ieee.org/document/7823109/ (Requires IEEE subscription.)

If you don't have an IEEE subscription, you can find the preprint version here for free on my Carnegie Mellon University web site: https://users.ece.cmu.edu/~koopman/pubs/koopman17_ITS_av_safety.pdf