If robots are built in which there are indeed unexpected variations in successive generations that must be weeded out to drive their evolution in a specific direction, or halt it at a certain point, that may indeed qualify as artificial life. If, on the other hand, robots are built which do what they are designed to do and there is no active selection required as successive generations evolve, that would not qualify as life.

Well, there would be error correction software included in any practical design, but I don't think it qualifies for what you're thinking of.

Basically, there will be some sort of software programs/data controlling the behavior of the machines. These would include the instructions on how to build more machines. This data will be stored somehow, perhaps on magnetic discs (like a hard drive); perhaps on solid state memory (like a flash drive); perhaps on punch cards. It will be stored somehow, and whatever method is used won't be absolutely perfect. A cosmic ray can flip a bit. The impact of a micrometeoroid could physically damage a section of the drive. Stuff happens.

To compensate for these possibilities, error correction software checks the integrity of the data using checksums and redundant copies of the data. If a bad block of data is detected, it is discarded and replaced/repaired/ignored.

If this error correction was not done, then there's the possibility that a "mutant" machine could be manufactured. Most "mutants" would not function, but there's a possibility that a mutation might copy itself and propogate.

With this error correction, the probability of a "mutant" becomes practically zero. It's easy to apply sufficient checksums to ensure there will be no mutation over trillions of years.