Update.

diff --git a/greed.py b/greed.py
index 20cef79..dc11d14 100755 (executable)
--- a/greed.py
+++ b/greed.py
@@ -77,6 +77,8 @@ def generate_episodes(nb, height=6, width=6, T=10, nb_walls=3, nb_coins=2):
         rnd = rnd * (1 - wall.clamp(max=1))
 
     rnd = torch.rand(nb, height, width)
+    rnd[:, 0, 0] = 0  # Do not put coin at the agent's starting
+    # position
     coins = torch.zeros(nb, T, height, width, dtype=torch.int64)
     rnd = rnd * (1 - wall.clamp(max=1))
     for k in range(nb_coins):