Safety Shouldn’t Happen by Accident
L2 autonomous tests demand highly qualified Safety Drivers to maintain the public’s confidence
As the trialling of autonomous vehicles becomes an ever more common sight on the public byways and highways, the role performed by a test driver is becoming the focus for a much greater level of scrutiny within the autonomous development community.
New proprietary standards for the role and function of the safety driver in urban trials are beginning to emerge with the clear intention to set a gold standard for the industry. Here’s what that taxonomy needs to include.
The starting point is nomenclature. It might sound trivial, but designating the role as ‘Safety Driver,’ not a ‘supervisor’ or a ‘test driver’ fundamentally alters the express purpose of the remit is safety and all other functional considerations for the job become secondary to this.
This is at odds with many of the large-scale autonomy trials common in the US . This ‘quantity-first’ focus stateside is largely conducted by gig economy operators supplied by staffing agencies that otherwise provide transit companies with bus drivers and tram conductors, whereas some smaller scale operations are employing highly skilled engineers as safety drivers.
There is definite merit in the volume approach — Waymo , for instance, are reporting reaching 20 million autonomous test miles. However by recording a far lower quantum of mileage at lower speeds to reflect the primary L2 use case in extra-urban environments. two goals are met by having a fully-qualified engineer at the helm of test vehicles unlike the volume operators.
Firstly, safety priorities can be enshrined far more intensively and secondly, the quality and the focus of the test data generated is far more useable and deployable and this means safe deployment can, ironically, be achieved more quickly.
Some companies only employ engineers as Safety Drivers. A case in point is StreetDrone whose Safety Driving team is led by Ross James, who joined the company after a four year period as a graduate safety engineer at Jaguar Land Rover. This approach brings invaluable OEM insight into the design and operation of their autonomous test and trial platforms as well as the company’s safety-case work.
As the safety case on the public roads is inherently linked to the design of the vehicle platform and software stack, a StreetDrone Safety Driver must have an in-depth knowledge of the underlying engineering allows them to trace every line of code through to every output the driving robot makes. The capability to interpret every output the autonomous system is making means the quality of the testing is far higher and our safety threshold is the very best it could possibly be.
Not only does Ross have a comprehensive understanding of the StreetDrone platform architecture, but also from his time at JLR, a full awareness of the underlying OEM design methodology. Like his colleagues, therefore, he has a comprehensive understanding of the entire engineering lifecycle inherent in delivering aa autonomous vehicle through trial and into scaling.
Ross was one of a team who recently completed a series of public road trials in central London, but when he’s not ‘in the saddle’ on trial, Ross provides a feedback loop to the software team to aid the continuous cycle of enhancements and leads the development of Operational Design Domains or ODDs, for StreetDrone’s forthcoming trials.
The ODDs define the parameters within which the trial will take place, for instance how the vehicle should contend with moving and static hazards, how it should operate in different weather conditions and therefore identify the junctures at which control should be handed over from the autonomous systems to the Safety Driver.
This amounts to a three-part prescription for Safety Drivers. Firstly, they must have an intimate knowledge of, and involvement in the platform engineering and the software. That way the Safety Driver can inherently read and respond to the behaviours of the autonomous systems during on trial, and of course contribute invaluable field-based insight back into the engineering iteration cycle.
Secondly, the Safety Drivers must be integral to the development of the ODD so they are participants in the creation of operating parameters rather than just recipients of a set of instructions.
And finally, train, train, train. Regardless of mileage and experience, Safety Drivers should go through routine, periodic training to ensure any variance from protocol is quickly identified and altered.
