Weaknesses in electromechanical assemblies will always remain. Batteries, optics, gyros and so forth.
Robotics that would be able to operate autonomous are pretty delicate in that each assy is so dependent upon another. If one assy is altered slightly, balance and mobility will cease to sync correctly. Every sensor is crucial to peak performance in operation.
Take for instance a sensor that registers the movement or full forward motion of a leg or an arm...whatever. That one sensor is vital to the proper synchronization of the forward/rear motion of the unit. Remove this sensor from the equation...robot moves no mo.
I could go on and on, but there will always be weaknesses.
Yeah, but remember, that my assumption was that these machines can self improve—something that humans can’t do.
Humans are kind of fragile too. We lose our two eyes and we are blind forever. A robot can have any number of eyes and these can be replaced if they go bad. Same deal with just about any non-healing body part. And, unlike a human, a robot can be healed in hours (or sooner) vs. days or weeks.
With nanobots, you have the capability to quickly evolve new structures from very small (cell sized) machines. The capacity of extremely rapid evolution is there.
Remember, we cannot explain why we are self aware. It is quite possible that a self-improving machine could create its own sentience by accident.