Merge pull request #26 from MIMUW-RL/jaco

keyboard agent for measuring human score
MIMUW-RL · Jun 27, 2022 · 8722c54 · 8722c54
2 parents 6e02281 + ab5dae3
commit 8722c54
Show file tree

Hide file tree

Showing 2 changed files with 48 additions and 16 deletions.
diff --git a/gym_space/__init__.py b/gym_space/__init__.py
@@ -83,7 +83,6 @@
     kwargs={
         "ship_steering": 1,
         "ship_moi": 0.01,
-        "reward_value": 0,
         "rad_penalty_C": 2,
         "numerator_C": 0.01,
         "act_penalty_C": 0.5,
@@ -102,7 +101,6 @@
     kwargs={
         "ship_steering": 1,
         "ship_moi": 0.01,
-        "reward_value": 0,
         "rad_penalty_C": 2,
         "numerator_C": 0.01,
         "act_penalty_C": 0.5,
@@ -121,7 +119,6 @@
     kwargs={
         "ship_steering": 1,
         "ship_moi": 0.01,
-        "reward_value": 0,
         "rad_penalty_C": 2,
         "numerator_C": 0.01,
         "act_penalty_C": 0.5,
@@ -140,7 +137,6 @@
     kwargs={
         "ship_steering": 1,
         "ship_moi": 0.01,
-        "reward_value": 0,
         "rad_penalty_C": 2,
         "numerator_C": 0.01,
         "act_penalty_C": 0.5,

diff --git a/keyboard_agent.py b/keyboard_agent.py
@@ -4,6 +4,8 @@
 import numpy as np
 
 if __name__ == "__main__":
+    total_rewards = []
+    episodes = 5
 
     gym.envs.register(
         id=f"KeplerDiscrete-v0",
@@ -25,18 +27,49 @@
     )
     #env = gym.make("KeplerDiscrete-v0")
 
+    gym.envs.register(
+        id="GoalDiscrete2-v0",
+        entry_point="gym_space.envs.goal:GoalDiscreteEnv",
+        max_episode_steps = 500,
+        kwargs={
+            "n_planets": 2,
+            "ship_steering": 1,
+            "ship_moi": 0.01,
+            "survival_reward_scale": 0.2,
+            "goal_vel_reward_scale": 5.0,
+            "safety_reward_scale": 10.0,
+            "goal_sparse_reward": 5.0,
+            "max_engine_force": 1,
+        },
+    )
     gym.envs.register(
         id="GoalDiscrete3-v0",
         entry_point="gym_space.envs.goal:GoalDiscreteEnv",
+        max_episode_steps = 500,
         kwargs={
             "n_planets": 3,
             "ship_steering": 1,
             "ship_moi": 0.01,
-            "survival_reward_scale": 0.0,
-            "goal_vel_reward_scale": 1.0,
-            "safety_reward_scale": 2.0,
-            "goal_sparse_reward": 10.0,
-            "max_engine_force": 0.4,
+            "survival_reward_scale": 0.2,
+            "goal_vel_reward_scale": 5.0,
+            "safety_reward_scale": 10.0,
+            "goal_sparse_reward": 5.0,
+            "max_engine_force": 1,
+        },
+    )
+    gym.envs.register(
+        id="GoalDiscrete4-v0",
+        entry_point="gym_space.envs.goal:GoalDiscreteEnv",
+        max_episode_steps = 500,
+        kwargs={
+            "n_planets": 4,
+            "ship_steering": 1,
+            "ship_moi": 0.01,
+            "survival_reward_scale": 0.2,
+            "goal_vel_reward_scale": 5.0,
+            "safety_reward_scale": 10.0,
+            "goal_sparse_reward": 5.0,
+            "max_engine_force": 1,
         },
     )
     env = gym.make(f"GoalDiscrete3-v0")
@@ -86,6 +119,7 @@ def key_release(key, mod):
     env.unwrapped.viewer.window.on_key_press = key_press
     env.unwrapped.viewer.window.on_key_release = key_release
 
+
     def rollout(env):
         global human_agent_action, human_wants_restart, human_sets_pause
         human_wants_restart = False
@@ -95,6 +129,7 @@ def rollout(env):
         total_reward = 0
         total_timesteps = 0
         k = 0
+
         while 1:
             if not skip:
                 # print("taking action {}".format(human_agent_action))
@@ -120,19 +155,20 @@ def rollout(env):
                 break
             while human_sets_pause:
                 env.render()
-                time.sleep(0.1)
+                time.sleep(0.01)
             k += 1
             time.sleep(0.1)
         print("END OF GAME! YOUR FINAL SCORE:")
+        total_rewards.append(total_reward)
         print("timesteps %i reward %0.2f" % (total_timesteps, total_reward))
         print(obser_max)
-        time.sleep(5)
+        time.sleep(2)
 
-    print("ACTIONS={}".format(ACTIONS))
-    print("Press left/right arrows (rotation) and space (engine)")
-    print("No keys pressed is taking action 0")
-
-    while 1:
+    for e in range(episodes):
         window_still_open = rollout(env)
         if window_still_open == False:
             break
+
+    print("HUMAN BASELINE SCORE:\n")
+    print(np.mean(total_rewards))
+    print(np.std(total_rewards))