Surgical 'robots' are pretty much entirely human operated cut-by-wire devices. "Waldos" of varying sophistication and shape.
Very handy because you can, say, mount the business end of the device on something a lot thinner and more flexible than a surgeon's wrist, and avoid having to crack the patient's entire chest open, or apply a filter between the input and the output, to allow the surgeon to make otherwise impossibly tiny motions.
To the best of my understanding, much of the remaining challenge is machine vision/sensing. Unlike assembly line robots, surgical bots can't make assumptions about product uniformity(indeed, if they have you cracked open for repair, abnormality is the only safe assumption, though even clinically normal people can vary considerably) and failure to correctly distinguish between tissue types or other visual mistakes can have unpleasant consequences.
In terms of pure steadiness, strength, or repeatability, humans are pretty screwed; but getting robots to stop fucking up magnificently when something unexpected happens has continued to be tricky.