A good exercise is to think of all of the computations required to point at something across the room. From a biomechanics perspective, there is a lot going on. Even if you just model each muscle in your arm/hand/finger as attached at two points and contracting, the force diagram of all the bones/joints quickly becomes exhausting.
The biomechanics of what humans do for basic locomotion is extremely complicated to model/recreate. While robots don’t need all the details (wheels, for example, make things easier) there is still a lot to deal with to make machines that can interact with our world.
It’s one thing to design an assembly-line robot that performs a single task. Making a robot that can do multiple useful things quickly explodes in complexity. There is steady progress in the field, but it’s slow.