We are very far away from a machine like that, and we may never be able to build such a roboter, like the vast majority of enginers who develope roboters and its software say so. But we certainly will have some amazing machines in future.
That's my department:
http://site.dutchrobocup.com/
Although I only worked on the vision subsystem. But I can confirm that the level we're working at is more like "how do we make sure it doesn't fall", or "how do we distinguish another robot from a shadow". It's nothing like Terminator (although our robot certainly looks cool), and definitely not the level where you need to worry about ethics. At least not about implementing "autonomous ethics" inside the robots. More about the ethics of how people use such robots.
I think robots might become like PCs in the future. They are a tool for people, not something you would like to let take autonomous decisions on a high level. For so far as it takes autonomous decisions, you want them to be easily understandable and predictable.
If American soldiers in Iraq can't distinguish terrorists from civilians, with all their training, and their human social abilities to guess the intentions of other human beings, then how will a robot ever be able to do so? When attempts are made to make a system do things like that, people will find out they are terribly unreliable.