New experiment

Author: sash-a

configs/flagrun.json

@@ -2,12 +2,12 @@
   "env": {
     "name": "AntFlagrunBulletEnv-v0",
     "max_steps": 1000,
-    "ant_env_rew_weight": 0,
+    "ant_env_rew_weight": 1,
     "path_rew_weight": 1,
     "kwargs": {
       "enclosed": true,
       "timeout": 100,
-      "size": 5
+      "size": 10
     }
   },
   "noise": {
@@ -40,10 +40,10 @@
     "use_pos": false
   },
   "general": {
-    "name": "flagrun-5eps_100timeout_5size",
+    "name": "flagrun-10eps_100timeout_10size-env_and_path_rew",
     "gens": 500,
-    "policies_per_gen": 4800,
-    "eps_per_policy": 5,
+    "policies_per_gen": 2400,
+    "eps_per_policy": 10,
     "n_policies": 1,
     "batch_size": 500,
     "seed": null,

run_saved.py

@@ -40,7 +40,11 @@ def run_saved(model: torch.nn.Module, env, steps):
     # # noinspection PyUnresolvedReferences
     # import pybullet_envs
 
-    e = gym.make(args.env, enclosed=True, timeout=-1).unwrapped
+    timeout = 200
+    world_size = 10
+    enclosed = True
+
+    e = gym.make(args.env, enclosed=enclosed, timeout=timeout, size=world_size).unwrapped
     e.mpi_common_rand = np.random.RandomState()
 
     e.render('human')