Car manufacturers like WAYMO and TESLA has been developing high value Autonomous (AD) systems that are said to be level 3.

AD systems are complex architectures in a feedforward chain : 
. perception
. Sensor fusion
. Situation understanding and holding
. Decision making
. Actuators

Such a feedforward chain is validated on crossings of use cases recorded in huge databases.
The problem is that it is not possible to validate a feedforward chain on a database and know what this feedforward chain will do our of the database. To make sure that an intelligent complex system can give a "good" answer in "unknown" situations, the best lead is to transform the feedforward chain into a feeback chain.
In order to do this one must chose a "measurable variable of performance".

NEXYAD proposed to usee the risk that "driver" (human or AD system) is currently taking. Indeed, they developed a module called SafetyNex that compute in real time at each moment this driving risk.

Then it is possible to check at each moment is the car is not taking too much risk, and if yes, then modify the decision of AD system (i.e. slightly slow down) in order to keep risk under a maximum accepted value :

schéma SafetyNx pour AD System


With such a scheme, it is not needed to modify the AD system, and the autonomous car can adapt its behaviour to unknown situations in order to keep risk at an acceptable value.

A good lead to go from level 3 to level 5 without developing expensive and complex systems.

Read more :