def pybullet_humanoid():
locals().update(default())
randomizer = (minitaur_env_randomizer.MinitaurEnvRandomizer())
env = 'HumanoidBulletEnv-v0'
max_length = 1000
steps = 3e8 # 300M
return locals()
def pybullet_duck_minitaur():
"""Configuration specific to minitaur_gym_env.MinitaurBulletDuckEnv class."""
locals().update(default())
randomizer = (minitaur_env_randomizer.MinitaurEnvRandomizer())
env = functools.partial(minitaur_gym_env.MinitaurBulletDuckEnv,
accurate_motor_model_enabled=True,
motor_overheat_protection=True,
pd_control_enabled=True,
env_randomizer=randomizer,
render=False)
max_length = 1000
steps = 3e7 # 30M
return locals()
def ResetPoseExample():
"""An example that the minitaur stands still using the reset pose."""
steps = 1000
randomizer = (minitaur_env_randomizer.MinitaurEnvRandomizer())
environment = minitaur_gym_env.MinitaurBulletEnv(
render=True,
leg_model_enabled=False,
motor_velocity_limit=np.inf,
pd_control_enabled=True,
accurate_motor_model_enabled=True,
motor_overheat_protection=True,
env_randomizer=randomizer,
hard_reset=False)
action = [math.pi / 2] * 8
for _ in range(steps):
_, _, done, _ = environment.step(action)
if done:
break
environment.reset()
def SinePolicyExample():
"""An example of minitaur walking with a sine gait."""
randomizer = (minitaur_env_randomizer.MinitaurEnvRandomizer())
environment = minitaur_gym_env.MinitaurBulletEnv(
render=True,
motor_velocity_limit=np.inf,
pd_control_enabled=True,
hard_reset=False,
env_randomizer=randomizer,
on_rack=False)
sum_reward = 0
steps = 20000
amplitude_1_bound = 0.1
amplitude_2_bound = 0.1
speed = 1
for step_counter in range(steps):
time_step = 0.01
t = step_counter * time_step
amplitude1 = amplitude_1_bound
amplitude2 = amplitude_2_bound
steering_amplitude = 0
if t < 10:
steering_amplitude = 0.1
elif t < 20:
steering_amplitude = -0.1
else:
steering_amplitude = 0
# Applying asymmetrical sine gaits to different legs can steer the minitaur.
a1 = math.sin(t * speed) * (amplitude1 + steering_amplitude)
a2 = math.sin(t * speed + math.pi) * (amplitude1 - steering_amplitude)
a3 = math.sin(t * speed) * amplitude2
a4 = math.sin(t * speed + math.pi) * amplitude2
action = [a1, a2, a2, a1, a3, a4, a4, a3]
_, reward, done, _ = environment.step(action)
sum_reward += reward
if done:
break
environment.reset()