Why autonomous vehicles need a large-systems approach to safety

The autonomous vehicle (AV) industry has yet to converge on a single method for measuring the maturity of AV technology with AV companies offering a range of different approaches for which metrics should be used to indicate system safety.

As AV projects evolve to full-scale deployments, the goal is to obtain widespread adoption such that the benefits of AVs (including improved road safety, mobility, and sustainability) can be realized on a broad scale. Research shows that public support for AV adoption hinges on how safe they are, rather than their economic impact; or privacy concerns stemming from the data they might collect.

In areas where testing is conducted, companies and local governments have communicated to the public that AVs are safe even though researchers agree that reliable data of crashes and near misses are “unable to provide information about safety”. Given this state of confusion, the public remains rightfully sceptical of self-driving vehicles.

Many of the existing solutions to measuring safety share one common feature: they are small-systems approaches; in other words, they evaluate AV safety through a singular lens, method or metric. Narrowly defining AV safety can simplify regulatory decisions but, at the same time, risks creating blind spots by neglecting the nascent, complex and non-deterministic nature of AV technology.

What is needed is an approach to safety that is both feasible and comprehensive

To address these challenges, AV Combinator has put forward the Autonomous Vehicle Modular Safety Suite white paper. AV Combinator is a group of researchers from Stanford University, founded in the Hacking 4 Defense programme, with a mission to close the communication gap between policy-makers and AV companies. Based on interviews with industry leaders and federal and local government officials, the white paper suggests a new approach to AV safety, outlining how it can be implemented for the improved understanding and assessment of AV safety in the US.

This new framework is a large-systems approach, in which each small-systems solution (such as simulation-based testing) is effectively a “module”. Using this approach, the user (AV developers, regulators or public-interest groups, for example) can choose from a pool of existing modules to develop their own proprietary “suite” based on their priorities, existing policies and risk tolerance.

A modular safety suite which is adaptable to a range of stakeholders, from AV Developers to the public, will greatly expedite AV deployment:

Compared to using only one safety metric or evaluating just one aspect of safety, creating a suite not only encourages users to evaluate AV safety in a comprehensive manner; consolidating these siloed small-systems solutions into one large system can help standardize how stakeholders communicate their safety requirements and comfort level to others in the system.

Creating a common safety language in such a way can make AV safety more accessible to the general public, while the holistic nature of the approach means that the public can be educated about its different facets. This will make it easier for individuals to understand the benefits and limitations of AVs as compared to human-driven cars, setting clear expectations which form the foundation of public trust and acceptance.

As more users adopt this large-systems approach and sharing becomes more widespread, there will be opportunities to consider correlations between specific modules and real-world safety performance. For instance, the adoption of one module might be found to be more strongly correlated with successful real-world driving tests as compared to other modules. This could indicate that the module is a relatively more accurate measure of AV safety and should be considered for integration into local and global safety standards.

The large-systems approach, therefore, thrives by embracing the complexities of this new technology. It provides all parties with the tools and information to enable the entire system to converge towards the best set of safety standards over time.

Additionally, a large-systems approach understands that autonomous driving is a “complex socio-technical innovation” at its core. Unlike a small-systems approach, a large-systems approach is able to adapt to changes and technological developments over time, which is especially important given how nascent the technology is. Users can swap out modules or add in new ones, as and when they are updated or developed.

This built-in flexibility means that a large-systems approach can also be applied in various markets. Given the range of nations developing their own approach to AV policy and safety assessment, modules can be added to safety suites according to the priorities of the regulator, existing regulations and safety culture of the region.

The Modular Safety Suite (MSS) is an implementable and robust example of a large-systems level approach. Developed by AV Combinator with the US AV governance system in mind, it is an initial proposal that can be iterated upon.

The MSS consists of six modules:

1. Traditional safety metrics: recording of vehicle crashes, injuries and fatalities;

2. Traditional crash testing: to evaluate the vehicle’s physical structure and ensure that the vehicle protects its occupants and other road-goers;

3. Aggregation of voluntary AV safety standards: consolidates and tracks how existing and future AV safety standards are used;

4. Proactive safety metrics: metrics that can be accurately gathered and assessed without removing the safety driver to measure the AV’s abilities in perception, prediction, planning and execution. An example of a proactive safety metric is the measurement of how consistent an AV is in obeying optimal positioning from other vehicles on the road, which can be done through scenario-based, simulated driving tests;

5. Comparative driving test: puts a human driver behind the wheel of an AV while the autonomous system runs in the background to specifically compare the performance of an AV with that of a human driver in a hyper-controlled setting;

6. Anonymized databases of best practices, safety metrics and driving scenarios: Trains and validates AV technology on a wider, more comprehensive dataset, while supporting the formation of data-driven safety standards.

The six modules in the MSS are split between lagging and leading measures. Lagging measures track only outcomes, such as a crash, once it has already occurred. Conversely, leading measures are proactive indicators that measure prevention efforts and can be observed and evaluated prior to a crash occurring, providing foresight to the technology’s performance prior to deployment. By encompassing both types of measures, the MSS intends to produce an output that gives a comprehensive view of AV safety.

Much like the modules themselves, the MSS will compete in the marketplace of safety systems. Federal, state and local regulators will select approaches from this marketplace to adopt, iterate and develop. This open marketplace will drive greater transparency in safety data and greater substantive safety for pedestrians and passengers alike.

Autonomous technology is expected to drastically improve the safety, sustainability, and mobility of our transportation systems. Acknowledging that creating a cohesive and inclusive approach to safety is the key to accelerating AV development, the large-systems approach offers a new way of thinking about AV safety.

For such an approach to be productive and useful, it must fulfill two conditions:

1. It must be technology-neutral. In other words, it cannot specify the use of a certain technology in creating the automated driving system. This aligns with the spirit of the current US Department of Transport policy that aims to provide an open regulatory environment to encourage innovation as a means of achieving safety;

2. Collaboration is critical. Automobile manufacturers, technology developers and relevant government agencies can create far more value working together than against each other. This means building mechanisms for policy development, knowledge sharing and establishing protocols.

If existing silos are to remain, the development of a cohesive safety framework, like the MSS, cannot move forward. A concerted effort by all stakeholders in the AV system is required to push through the status quo for the adoption of this new approach. Otherwise, developing trusted safety measures and gaining public acceptance for AVs will likely continue to be an uphill battle.

AV Combinator is working towards exploring how exactly this large-systems approach can be applied to different jurisdictions and cities. To learn more about the MSS, read AV Combinator’s Autonomous Vehicle Modular Safety Suite White Paper.