Driving behavior has been assessed subjectively since before the advent of the automobile; occupants and observers of an automobile have a subjective notion of the driving behavior for that vehicle. However, until now — after more than a century’s worth of automobile driving — there are no agreed standards to objectively assess driving behavior, with the exception of (backwards-looking) historical collision statistics.

Remarkably, even the U.S. Department of Motor Vehicles Driving Performance Evaluations are subjective, dependent on the examiner’s perception and the DMV’s location.

Nothing has changed for a long, long time — and now it must.

Today, with our roadways shared by human and non-human drivers, subjective assessments of safety are no longer sufficient. And frankly, regardless of their applicability to human drivers, safety indicators such as historical collision statistics are not relevant nor appropriate for assessing non-human drivers.

There is a real need for a safety assurance toolchain that can objectively assess human drivers and non-human drivers sharing the same roads. Before autonomous vehicles are widely deployed on our roadways, key mobility-related stakeholders must answer these two crucial questions:

• How safe is safe-enough?
• And how do we measure it?

Developers of autonomous systems, as well as potential deployers, customers, and regulators, are all wondering when Autonomous Driving Systems (ADS) will be safe enough to deploy. Currently, there is no agreement, less a method, to answer this question. Existing approaches, such as historical collision statistics, Time-to-Collision (TTC), Responsibility-Sensitive Safety (RSS), Instantaneous Safety Metric (ISM), Inertial Measurement Unit (IMU) data, and disengagements do not provide the information and context required to provide a measurement of safety that will be useful to these stakeholders.

To be effective, a safety metric must be:

• fully objective, even agnostic
• applicable to human and non-human drivers alike
• a leading indicator, not relying on rare historical events of human-driven vehicles or actual AV incidents
• easily and consistently measured with inexpensive, readily available sensors
• independent, able to be measured and assessed just as easily by regulators or third-party evaluators as by the system developers themselves

The Streetscope Collision Hazard Measure (SHM™)

Streetscope offers an objective and general measurement of collision hazard, calculated by using movement data of all traffic objects. The state-of-the-art Streetscope Collision Hazard Measure (SHM™) is a continuous calculation of relative hazard that quantifies the degree of near-miss for each pair of traffic object interactions occurring between all the traffic objects in any traffic scenario.

The Streetscope Collision Hazard Measure™ is:

• Leading
• Quantitative
• Free of Assumptions
• Continuous
• Independent
• Repeatable
• Monotonic
• Objective
• Computable
• Scalable

Perhaps most importantly, the Streetscope Collision Hazard Measure is applicable to human drivers and non-human drivers alike and is also validated across all modes of AV system development: on road, on track, in simulation, and for safety assurance in operation.

Our CTO, Erik Antonsson, published a white paper on how SHM™ overcomes the limitations of existing safety metrics (TTC, RSS, ISM, IMU) and provides an independent leading indication of safety.

Streetscope Safety Assurance Toolchain

Streetscope’s safety assurance toolchain is anchored by the SHM™, which helps to answer the two questions:

How safe is safe-enough? And how do we measure it?

Streetscope’s toolchain is sensor-agnostic and collects data on the position of all traffic objects, on a frame-by-frame basis, using various data ingestion methods to calculate SHM™. To be useful in as many contexts as possible, Streetscope has also built a cost-effective video ingestion system for widely deployed high-rate sensors such as fixed-infrastructure cameras or GPS-enabled dash-cams.

The resultant Safety Performance Indicators provide an aggregate indication of hazard based on near- misses, correlating to risk associated with vehicles, environmental conditions, or infrastructure, whether under piloted or autonomous control. Controlling certain factors (e.g., test routes, environments) allows for direct comparison of safety performance between vehicles or control strategies.

Streetscope Hazard Assessment

The Streetscope Safety Assurance Toolchain answers important questions for AV developers and deployers, transportation infrastructure companies, insurers, and regulators:

• Risk assessment of different versions of AV Stack and/or human drivers
• Risk assessment of Street Corridors and Intersections to reduce collisions
• Measure incremental safety performance of each improved version of AV stack
• Measure AV stack safety performance on existing roadways
• Risk assessment of Operational Design Domain (ODD) for new AV service deployment/operation
• Driver (human/AV) Training using high-hazard interactions (near-miss scenarios) database
• Driver and Environmental Safety scores for Insurance underwriting

Streetscope produces life-saving, actionable insights that improve the safety of our streets and accelerates the deployment of safe autonomous mobility with the assurance of an objective measure of safety.

‍

\\\

‍

Image by Streetscope