The easiest part of the project will be the animation core, and should be tackled first. Hopefully, the routines will be generic enough and documented enough to be used in other projects which might even be entirely unrelated to this project.
The second stage will be the physics core. In order to be useful, this stage has to be taken above and beyond the simple collision-response system used in video games. The project has to realistically simulate the workings of complex machinery. Finite Element Material Simulation may be required. Physics will control animation. This part will need to also simulate electronic circuits, even if only at an abstracted level.
The third stage will be artifical intelligence. This is partly to make the project useful earlier on in its development. The AI routines only need to be sophisticated enough to replicate lower cognitive functions: enough for a simulated android/gynoid to keep balance and navigate simple obstacles. This stage may be developed further if required and should pose no real difficulty if the architecture remains modular. AI has control over physics, but obviously there has to be two-way communication between the observer and the subject.
The fourth stage will be simulation of a biological system - a human - at an appropriate level. Further levels will be added as needed. At this stage, distributed computing will definitely be required.
The fifth stage, if required, will replicate realistic environments for the simulated cybernetics system to interact with.
C++ is the current industry standard, but to aid future-proofing, compatibility and distributed computing capabilities, Java should be used for some components. Considering how long it may take for this project to reach maturity, it may be deemed prudent to develop a Java version in tandem with the C++ version. In fact, it might be preferable to not use C++ at all.