Wouldn't using actual nerve impulses be very difficult require surgery etc and be really expensive.
The idea behind this is its relatively cheap
It depends.
If a person with a fully functional brain is to use nerve impulses to
control an artificial arm, then not really. I think it might be necessary to have surgery done to keep the nerves working, but I'm not sure about that. The arm itself though, definitely not. It's possible to detect nerve signals from outside, so a band of sensors on the remaining natural part of the arm will be fine for detecting the nerve signals and a quite low-power computer will be fine for interpreting them and converting them into signals to control the motors in the artificial arm. Most of the work is still being done by the person's brain - that calculates required positioning, strength, duration, etc and generates the required signals.
If a person with a fully functional brain is to use nerve impulses to
feel with an artificial arm, then yes. It requires specialised surgery and complex equipment and specialised configuration (and therefore serious expense) to get artificial signals from sensors on the artificial arms into the person's brain as signals from their arm. It can be done and it has been done, but it's difficult and expensive to do. The working setup I saw involved a specialised surgeon re-routing nerves from a person's fingertips up to their shoulder during an amputation of their arm and then additional hardware to stimulate those rerouted nerves according to signals from the sensors on the fingertips of the artificial arm. The patient's brain adapted to the change in input very quickly and integrated it into their proprioception, so they perceived it as sensory information from their own fingertips despite the fact that it was sensors on an artificial arm sending signals to hardware to stimulate nerves on their shoulder. It's a one-off prototype (to the extent that the hardware stimulating the rerouted nerves was being held in place by the surgeon because there wasn't any sort of mounting yet) but it proved that the approach works. The patient obviously knew what was really happening, but that didn't stop it working. The human brain is versatile enough to work around the change and integrate the new signals.
If the body control parts of the person's brain aren't working correctly, it's a different matter entirely. That would probably require a powered exoskeleton controlled by conscious thought, which would probably not be good enough anyway. Conscious thought is too slow.