Finalize the RealTimeRuntime class

[diegoferigo]

We should finalize the implementation of the RealTimeRuntime python class, that allows stepping the Tasks ensuring that the agent rate is respected.

It should be fairly easy taking inspiration of the existing logic of the PyBulletRuntime:

gym-ignition/gym_ignition/runtimes/pybullet_runtime.py

Lines 182 to 207 in 960a856

	def _enforce_rtf(self):
	if self._now is not None:
	# Compute the actual elapsed time from last cycle
	elapsed = time.time() - self._now
	# Nominal timestep with the desired RTF
	expected_dt = 1.0 / self.agent_rate / self._rtf
	# Compute the amount of sleep time to match the desired RTF
	sleep_amount = expected_dt - elapsed - self._bias
	# Sleep if needed
	real_sleep = 0
	if sleep_amount > 0:
	now = time.time()
	time.sleep(sleep_amount)
	real_sleep = time.time() - now
	# We use a low-filtered bias to compensate delays due to code running outside
	# the step method
	self._bias = \
	0.01 * (real_sleep - np.max([sleep_amount, 0.0])) +\
	0.99 * self._bias
	# Update the time for the next cycle
	self._now = time.time()