added more example files for pure dep exploration

deprl/dep_controller.py

@@ -11,7 +11,7 @@
 class DEP:
     """
     DEP Implementation from Der et al.(2015).
-    Jax is used instead of numpy to speed up computation, GPU strongly
+    PyTorch is used instead of numpy to speed up computation, GPU strongly
     recommended.
     In the future, proper JAX features such as jit, vmap, etc
     should be used.

examples/example_only_dep_myosuite.py

@@ -0,0 +1,25 @@
+import gym
+import myosuite
+import time
+from deprl import env_wrappers
+from deprl.dep_controller import DEP
+
+
+
+env = gym.make('myoLegWalk-v0')
+env = env_wrappers.GymWrapper(env)
+
+# You can also use SconeWrapper for Wrapper
+# env = env_wrappers.SconeWrapper(env)
+dep = DEP()
+dep.initialize(env.observation_space, env.action_space)
+
+env.reset()
+for i in range(1000):
+    action = dep.step(env.muscle_lengths())[0,:]
+    print(action.shape)
+    next_state, reward, done, _ = env.step(action)
+    time.sleep(0.01)
+    env.mj_render()
+
+

examples/example_only_dep_scone.py

@@ -0,0 +1,49 @@
+import gym
+import sconegym
+from deprl import env_wrappers
+from deprl.dep_controller import DEP
+
+
+# create the sconegym env
+env = gym.make('sconewalk_h2190-v1')
+
+# apply wrapper to environment
+env = env_wrappers.SconeWrapper(env)
+
+# create DEP, parameters are loaded from default path
+dep = DEP()
+
+# give DEP obs and action space to create right dimensions
+dep.initialize(env.observation_space, env.action_space)
+
+env.seed(0)
+
+for ep in range(5):
+    if ep % 1 == 0:
+        env.store_next_episode()  # Store results of every Nth episode
+
+    ep_steps = 0
+    ep_tot_reward = 0
+    state = env.reset()
+
+    while True:
+        # samples random action
+        action = dep.step(env.muscle_lengths())[0,:]
+        # applies action and advances environment by one step
+        state, reward, done, info = env.step(action)
+
+        ep_steps += 1
+        ep_tot_reward += reward
+
+        # check if done
+        if done or (ep_steps >= 1000):
+            print(
+                f"Episode {ep} ending; steps={ep_steps}; reward={ep_tot_reward:0.3f}; \
+                com={env.model.com_pos()}"
+            )
+            env.write_now()
+            env.reset()
+            break
+
+env.close()
+