TL;DR; Quick and effective extraction of the long-tail of real but rare scenarios are critical to quickly getting AV/ADAS off-the-ground into the real-world. Furthermore, by leveraging insights on the variabilities of such long-tail rare scenarios, scaling them would get easier and affordable. Rydesafely’s three products – Ryde-EuroNCAP, Ryde-ADAS, and Ryde-Autonomy, enables such insight extraction for EuroNCAP, ADAS and higher levels of autonomy respectively.

Rare but realistic scenarios are the impedances to AV/ADAS deployment

I am sure we have all been in a traffic situation where someone has driven unexpectedly. Some examples of these kind of situations could be

• speeding up to cross an intersection right before the traffic goes red,
• violating traffic-flow direction
• blocking traffic at an intersection,
• multiple lane changes to take a highway exit, etc.

Turns out these are the exact kind of scenarios that is blocking AV/ADAS deployment to any Operational Design Domains (ODD). While some are easy to imagine, many are unfathomably complex and surprising. The long-tail of such rare but realistic events on roads could be infinite due to varied factors such as weather, road conditions, time of day etc.

What makes it even a harder problem is such rare events might also have a localised attribute that makes the AV/ADAS scaling strategy to new ODDs (eg; new geographies) a formidable problem. By the way, we have covered a few ideas around scaling in another post here.

Abundance driving data is a double-edged sword

The long-tails of rare events, their variabilities have warranted digital twins and high-fidelity simulators but the problem still remains. It still seems like sourcing driving data is the best way going forward as clues about what long-tail to validate and build a system for can be extracted from traffic insights. However, driving data racks up very quickly to colossal sizes. This makes the whole purpose of finding the long-tail of rare events go into a tail-spin. Hence, it is pivotal that these rare but realistic events are found automatically, with accuracy and speed, and with several coverage arguments for an easy 10k feet view.

Rydesafely’s Driving Reasoning Engine to the rescue

Rydesafely’s Driving Reasoning Engine has the ability to extract the rare long-tail scenarios from colossal amounts of data automatically in 3 steps. It further goes a step ahead to offer two-fold coverage arguments

• data-representativeness for the entire ODD of interest,
• scenario type coverage against all possible occurrences in the ODD of interest.

You can see some rare but important scenarios extracted by Rydesafely’s Driving Reasoning Engine.

Rydesafely’s three products – Ryde-EuroNCAP, Ryde-ADAS, and Ryde-Autonomy are solving this problem for EuroNCAP, ADAS and higher levels of autonomy respectively. These products are built on top of the Driving Reasoning Engine and geared towards addressing the long-tails for specific use-cases. More information can be found on Rydesafely’s website. Very high-level differences of the three products are highlighted below.

Further information is available on request including demonstrations and a trail software. Either use this link to submit a request or send an email to

• sanjay@rydesafely.com
• tom@rydesafely.com.

This article is also posted in Rydesafely’s blog here.