Presented by Federico Giaimo.

Faults in a complex system like a car can be very dangerous. Especially nowadays with self-driving cars. The general way of dealing with faults in the automotive is by recalling the cars, and update them. However, this is extremely costly. So, can continuous experimentation help to improve quality, and thus less recalls?

To allow for this, the software architecture should support this. Another problem is that such systems have limited resources available. They consider three strategies: parallel execution (just have two processors), serial execution (which can be used if not all of the time share is being used by the modules) and down-sampling (once every n executions, the experimental module is used, instead of the production module). The latter can be a problem if the system is stateful. Also, external conditions may prevent the latter. Therefore, they identified relevant criteria that must be fulfilled to allow down-sampling.

ECSA 2017: Considerations about Continuous Experimentation for Resource-Constrained Platforms in Self-Driving Vehicles