class Evaluator(object):
def __init__(self):
self.world = WorldManager()
self.uni = Unicycle()
self.uni.add_to_manager(self.world)
def evaluate(self, brain):
fitness = 0
self.time = 0
self.uni.reset(0.05)
while (self.time < MAX_EVAL_TIME):
self.step(brain)
in_pitch_threshold = abs(self.uni.pitch()) < PITCH_SCORE_ANG
in_roll_threshold = abs(self.uni.roll()) < ROLL_SCORE_ANG
in_yaw_threshold = abs(self.uni.yaw()) < YAW_SCORE_ANG
in_pos_threshold = abs(self.uni.x()) < 2.5 and abs(self.uni.z()) < 2.5
if in_pitch_threshold and in_roll_threshold and in_yaw_threshold and in_pos_threshold:
fitness += 0.4 * DT * dirac_delta(self.uni.yaw(), 0.8)
fitness += 0.3 * DT * dirac_delta(self.uni.kinetic_energy(), 0.01)
fitness += 0.3 * DT * dirac_delta(abs(self.uni.x()) + abs(self.uni.z()))
else:
break # Failure. No need to evaluate further.
return fitness
def step(self, brain):
input = [
self.uni.pitch(),
self.uni.pitch_velocity(),
self.uni.roll(),
self.uni.roll_velocity(),
]
output = brain.feed(input)
self.power_used = abs(output[0]) + abs(output[1])
# (-1 if self.uni.roll() < 0 else 1)
# (-1 if self.uni.pitch() < 0 else 1)
self.uni.apply_drive_impulse(
self.uni.drive_impulse * output[0] #* random.gauss(1, 0.2)
)
self.uni.apply_wheel_impulse(
self.uni.wheel_impulse * output[1] #* random.gauss(1, 0.2)
)
self.world.step_simulation(DT)
self.uni.compute_state(DT)
self.time += DT
def __init__(self):
self.world = WorldManager()
self.uni = Unicycle()
self.uni.add_to_manager(self.world)
