Despite some bold predictions for autonomous vehicles (AVs) of late, enthusiasm for further development has stalled. Foretellix , an Israeli based start-up, believes that verifying the safety of an AV through a miles-driven approach is time-consuming and costly. Their focus is on the quality of coverage and offer a new language for AV developers to sustain development. To learn more, we spoke to CEO and co-founder Ziv Binyamini.
The most challenging barrier to vehicle autonomy is ensuring the autonomous vehicle behaves properly under all possible conditions. What is your approach to safety verification of ADAS systems and AVs?
We founded Foretellix after many years working on coverage driven verification for chips in the semiconductor industry. This approach is now standard in that industry. We adapted this approach to the safety verification of ADAS systems and AVs.
The number of scenarios AVs need to successfully contend with, and the different permutations and variables of them, is almost limitless. Current metrics used in the industry are miles and disengagements: AVs are subjected to millions of miles of physical driving. Engineers then record each disengagement, the occasions when the driver has to take back control of the vehicle. Industry experts agree that these metrics are flawed.
In contrast, our approach is based on 'Quality of Coverage', rather than the 'Quantity of Miles' approach. We verify AV safety by using objective metrics and a proven coverage driven approach to ensure AVs can successfully deal with the huge number of scenarios possible on the open roads.
How does your quality of coverage approach work?
Our approach relies on a scenario specification language used by AV developers to specify the high-level scenarios. We then use automation to generate many variants and combinations of concrete scenarios and execute them on various testing platforms. We automatically monitor all driving tests, collecting coverage information and aggregating it into a single metric dashboard.
By automating the process of identifying and specifying these 'edge case' scenarios, we enable AV testing to become much more efficient and quantifiable.
What does your approach mean for consumers, regulators and insurance companies?
Today, the enthusiasm for AVs has stalled because proving they are safe is still yet to be achieved.
Today, the enthusiasm for AVs has stalled because proving they are safe is still yet to be achieved. Our coverage driven approach will help all parties gain the confidence in autonomous vehicle safety, and in turn, speed up the deployment of AVs on our roads.
Our coverage driven verification approach allows AV developers to successfully exercise the scenarios critical for AV safety and extract the metrics to prove it. These metrics will cover millions of driving scenarios, including 'edge case' scenarios that are crucial in helping developers, insurance companies and regulators understand AV safety.
Those metrics can also contribute to the establishment of targets that AV developers will need to meet. Objective metrics can be used and translated into a ratings-based system, similar to that used by current crash test organisations like EuroNCAP. A star-based system that demonstrates to consumers, regulators and insurance companies how safe an AV is will aid transparency, trust and confidence in the underlying technology.
To what extent can your approach accelerate the development of AVs?
The Quantity of Miles testing approach is hugely inefficient, costs a lot of money and takes up a lot of time. In addition to the need to physically drive hundreds of thousands of miles, AVs often need to be shipped to many different locations, so engineers can test them in different environments, including in different weather conditions. When a small fix is required, engineers need to go out and reproduce all of the scenarios again, which takes up even more time and resources.
We believe our approach will drastically speed up the deployment of AVs on the roads. By using objective metrics and utilising a coverage driven verification approach, AV developers can automatically recreate all of these different scenarios and validate their fixes overnight. This significantly reduces the need for endless repeated physical driving.
Is there a common industry approach to defining scenarios that AVs will face?
One of the biggest challenges in bringing AVs to market is that there isn't a common industry approach in defining these scenarios, or to the metrics used to present the safety case. Foretellix has developed a Measurable Scenario Description Language (M-SDL) – a human readable language to help developers articulate these scenarios. By using the same language, stakeholders, including developers, regulators and insurers, will be able to use a single, common scenario description language standard when assessing different AVs from different OEMs.
Last year, we opened up M-SDL to the wider AV ecosystem to enable OEMs and tech developers to assess its viability as a common scenario specification and provide feedback. Over 400 engineers from 150 companies, regulatory agencies and universities have since downloaded M-SDL. We have received very positive & valuable feedback.
The wider AV industry recognizes the need for a standard readable scenario description language and a common approach to testing and defining scenarios. The Association for Standardization of Automation and Measuring Systems (ASAM ), of which many global OEMs and AV developers are members, took on itself to implement a single standard for AV safety verification for scenario specification and formed a committee called OpenSCENARIO 2.0, for the creation of this standard. Foretellix contributed the M-SDL language concepts to ASAM OpenSCENARIO 2.0 and we are actively working with the committee to define the standard.
How does your Measurable Scenario Description Language work?
M-SDL enables intelligent and scalable automation, analytics and metrics. It enables two significant breakthroughs to help manage the incredibly complex challenge of identifying and verifying the millions of driving scenarios AVs will need to meet. The 'One to Many' breakthrough allows our automation to take 'One' scenario specification and generate 'Many' scenario variants and combinations across multiple environments and locations. The 'Many to One' breakthrough allow results from 'Many' tests executed across many testing platforms and locations to be aggregated into 'One' coverage dashboard and metric. These testing platforms include physical test track testing; simulation testing, X-in-the-loop testing and on-road testing.
What's next for Foretellix?
We will continue to collaborate and work with OEMs and AV developers to overcome the challenges in implementing Level 2, 3 and 4 autonomy. We will also continue to work closely with our AV ecosystem partners in integrating our software platform with their software platforms. Finally, we will continue working closely with ASAM, to define the OpenSCENARIO 2.0 standard for scenario specification.