added the learning algorithm from RL-lab

examples/pybullet/gym/envs/bullet/cartpole_bullet.py

@@ -24,13 +24,8 @@ class CartPoleBulletEnv(gym.Env):
 
   def __init__(self):
     # start the bullet physics server
-#    cmdStartBulletServer=['/Users/jietan/Projects/bullet3/build_cmake_python3/examples/SharedMemory/App_SharedMemoryPhysics_GUI']
-#    subprocess.Popen(cmdStartBulletServer)
-    # wait to make sure that the physics server is ready
-#    time.sleep(1)
-    # connect to the physics server
-#    p.connect(p.SHARED_MEMORY)
     p.connect(p.GUI)
+#    p.connect(p.DIRECT)
     observation_high = np.array([
           np.finfo(np.float32).max,
           np.finfo(np.float32).max,

examples/pybullet/gym/minitaur_bullet_gym_example.py

@@ -1,27 +0,0 @@
-import gym
-import numpy as np
-import math
-
-from envs.bullet.minitaur_bullet import MinitaurBulletEnv
-
-def main():
-    environment = gym.make('MinitaurBulletEnv-v0')
-    sum_reward = 0
-    steps = 1000
-    amplitude = 0.5
-    speed = 0.3
-
-    for stepCounter in range(steps):
-      a1 = math.sin(stepCounter*speed)*amplitude
-      a2 = math.sin(stepCounter*speed+3.14)*amplitude
-      action = [a1, 0, a2, 0, 0, a1, 0, a2]
-      state, reward, done, info = environment.step(action)
-      sum_reward += reward
-      print(state)
-      if done:
-        environment.reset()
-    average_reward = sum_reward / steps
-    print("avg reward: ", average_reward)
-
-
-main()

examples/pybullet/gym/trpo_cartpole_bullet_gym.py

@@ -0,0 +1,51 @@
+from envs.bullet.cartpole_bullet import CartPoleBulletEnv
+from rllab.algos.trpo import TRPO
+from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline
+from rllab.envs.gym_env import GymEnv
+from rllab.envs.normalized_env import normalize
+from rllab.misc.instrument import stub, run_experiment_lite
+from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy
+import subprocess
+import time
+
+stub(globals())
+
+env = normalize(GymEnv("CartPoleBulletEnv-v0"))
+
+policy = GaussianMLPPolicy(
+    env_spec=env.spec,
+    # The neural network policy should have two hidden layers, each with 32 hidden units.
+    hidden_sizes=(8,)
+)
+
+baseline = LinearFeatureBaseline(env_spec=env.spec)
+
+algo = TRPO(
+    env=env,
+    policy=policy,
+    baseline=baseline,
+    batch_size=5000,
+    max_path_length=env.horizon,
+    n_itr=50,
+    discount=0.999,
+    step_size=0.01,
+    # Uncomment both lines (this and the plot parameter below) to enable plotting
+#    plot=True,
+)
+
+#cmdStartBulletServer=['~/Projects/rllab/bullet_examples/run_physics_server.sh']
+#subprocess.Popen(cmdStartBulletServer, shell=True)
+#time.sleep(1)
+
+
+run_experiment_lite(
+    algo.train(),
+    # Number of parallel workers for sampling
+    n_parallel=1,
+    # Only keep the snapshot parameters for the last iteration
+    snapshot_mode="last",
+    # Specifies the seed for the experiment. If this is not provided, a random seed
+    # will be used
+    seed=1,
+ #   plot=True,
+)

examples/pybullet/gym/trpo_tf_cartpole_bullet_gym.py

@@ -0,0 +1,48 @@
+from envs.bullet.cartpole_bullet import CartPoleBulletEnv
+from sandbox.rocky.tf.algos.trpo import TRPO
+from sandbox.rocky.tf.policies.gaussian_mlp_policy import GaussianMLPPolicy
+from sandbox.rocky.tf.envs.base import TfEnv
+
+from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline
+from rllab.envs.gym_env import GymEnv
+from rllab.envs.normalized_env import normalize
+from rllab.misc.instrument import stub, run_experiment_lite
+
+stub(globals())
+
+env = TfEnv(normalize(GymEnv("CartPoleBulletEnv-v0")))
+
+policy = GaussianMLPPolicy(
+    name = "tf_gaussian_mlp",
+    env_spec=env.spec,
+    # The neural network policy should have two hidden layers, each with 32 hidden units.
+    hidden_sizes=(8,)
+)
+
+baseline = LinearFeatureBaseline(env_spec=env.spec)
+
+algo = TRPO(
+    env=env,
+    policy=policy,
+    baseline=baseline,
+    batch_size=5000,
+    max_path_length=env.horizon,
+    n_itr=50,
+    discount=0.999,
+    step_size=0.01,
+    force_batch_sampler=True,
+    # Uncomment both lines (this and the plot parameter below) to enable plotting
+    #plot=True,
+)
+
+run_experiment_lite(
+    algo.train(),
+    # Number of parallel workers for sampling
+    n_parallel=1,
+    # Only keep the snapshot parameters for the last iteration
+    snapshot_mode="last",
+    # Specifies the seed for the experiment. If this is not provided, a random seed
+    # will be used
+    seed=1,
+    #plot=True,
+)