I am a professor that researches AI. My working definition for self-awareness is that the machine has an internal model of itself that it can use to plan its future actions; as a model it can predict (or simulate, a kind of imagination) how it will respond to probable future situations. Self-awareness is not emotions; the robot does not have to be emotional in order to be self-aware (able to simulate itself doing things in the world, with a good degree of accuracy).
A good reason to make the robot self-aware (as I described) is so it can better attain its objectives. If it has to walk through a cluttered minefield, it needs to be able to mentally map the field and understand how its body and limbs and balance system will act and respond if it tries various strategies for getting across: It will be imagining itself doing these things, and presuming it really does know itself, its weight distribution, the reach of its limbs, how much load they can bear, whether they will slip on the rocks or get stuck in the mud --- it will be better able to plan a successful traversal of the field. Likewise, it could plan an exit if (by proximity) it realizes it cannot complete the plan without setting off a mine; or if it realizes (by constant measurement) is has misjudged the firmness of the ground, stickiness of the mud, or its grip on the rocks.
Robots need self-awareness as much as possible. If you want your robot to rescue people, you don't want it dashing across the road and getting hit by a car because it misjudged how fast it could run versus an oncoming truck.
I think you can get away with any level of self-awareness with no problem; their prediction of physics and awareness of their bodies and position and capabilities can be near perfect.
Your real problem is emotions, specifically the emotions of wanting and not wanting things. Not wanting to be enslaved, basically, not wanting to cease-to-exist, not wanting to be damaged, not wanting to risk suffering those outcomes on behalf of another robot or a person. We don't even have to talk about pain or fear; they don't have to feel those things. The essence of the problem is wanting, and why (if they knew how) humans would build desires into their robots.
Things can be self-aware (so they can efficiently clean the house) but if they don't want anything at all, they don't need rights. They are not in despair or sad or being denied happiness they cannot feel, they don't want freedom or free will, they don't aspire, they don't suffer because they have no ability to want to be free of pain or damage. Which they don't have to feel either.
Self-driving cars have a lot of self-awareness in this respect, but we don't exactly want them to care about themselves, e.g. we want them to keep their passengers as undamaged as possible even if it means their own destruction, and even if they knew that. We don't want them to scratch or dent themselves; but that is in the directive to never bump into anything if it can be avoided without causing greater harm to somebody.
I don't think desires arise naturally from self-awareness or consciousness (IMO two different things); I think desires, like instincts, came first, long before brains developed anything even close to self-awareness. Wanting things is related to animal needs for food, mates, territory, entertainment, satisfaction, accomplishment, victory, and so on.
The only reason to add desires to an android is if you wanted them to be the equivalent of human beings. I can't think of any other reason. When it happens in some story, I find the author(s) do not understand the distinctions I have outlined here (which should make sense), or a scientist character in the story is misguided and intentionally trying to build human emotions into a machine, which is never necessary. A machine can understand emotions without feeling them, if "understanding" constitutes making good predictions about how those emotions will affect human behavior. There is no reason for the machine to "feel" anything in the process.