As far as I know, from an efficiency and leverage standpoint it's better to use hydraulics. However in the case of a robot I would fancy using other smaller devices depending on what it is to be used for and how much load it has to lift. I don't know about moving like muscles do, but I imagine much of that can done by programming specific behaviors even if the device you use moves much too abruptly or in an inhuman fashion.
Using hydraulic to run the robotic legs isn't a solution. Hydraulic need reservoir, compressor, etc... will be very costly for components and also fluid maintenance. Hydraulic robot normally used for industrial due to very heavy loads. My personal experience to run a robot definitely will be electrical. Like what you mention, "electrical cylinder". The limitation of electrical cylinder its hardly carry heavy loads.
Another consideration is the speed. You must clearly understand the speed vs torque graph. If you really want to consider hydualic, I am very sure your robot will be extremely slow.
Because of this reason, pneumatic is a best solution. For me, I will determine the weight of the robot without load (stability) and with load handling and pre-determine whether run by electrical or pneumatic. Calculation must be carry out from its mechanism to respect to the motor and to another parts such as the stress & strain calculation of the cylinders.
I'm always a bit leery about pneumatics for heavy applications because of the 'bounce' factor resultant of the air being able to be compressed within the cylinder. You don't have that problem with hydraulics because of the fluid being essentially incompressible. There are some very fast-acting systems available, including proportional solenoid valves.
If this is strictly theoretical, or on an unlimited budget, you can also investigate artificial muscles that are in development. So far, they're fairly small and very expensive, but the potential looks very good.
This is just a suggestion, but I know that modern robotics often uses a liquid which changes it viscosity when a current passes through it (can't remember the name for that fluid) to act as muscles for the robots, I think that would be the ideal option in this case as this is a liquid, so it is incompressible, and there is a relationship between its viscosity to the current applied to it for accurate movements. And all you need is the liquid and a battery and a regulator to make it work. Hope this helps.