It seems like "Self-Driving" to the authors means simply that it has a query optimizer using ML.

I would counter that idea by suggesting that a more foolproof database would also incorporate truly solid auto-scaling, and global replication, so that one can just throw immense amounts of data into tables, while retaining fast CRUD operations worldwide, without any DBA support required. A little bit like the direction CockroachDB and TiDB are headed, perhaps.