From 61b289feefba9f0c8a4aa1aed57d3ac1bfadc9a1 Mon Sep 17 00:00:00 2001
From: Erik Nygren <erik.nygren@sbb.ch>
Date: Sat, 5 Oct 2019 09:31:00 -0400
Subject: [PATCH] updated single agent navigation to work with new env
---
torch_training/training_navigation.py | 7 +++----
utils/observation_utils.py | 2 +-
2 files changed, 4 insertions(+), 5 deletions(-)
diff --git a/torch_training/training_navigation.py b/torch_training/training_navigation.py
index ad512c6..3a61d1f 100644
--- a/torch_training/training_navigation.py
+++ b/torch_training/training_navigation.py
@@ -38,8 +38,7 @@ def main(argv):
x_dim = 20
y_dim = 20
n_agents = 1
- n_goals = 5
- min_dist = 5
+
# Use a the malfunction generator to break agents from time to time
stochastic_data = {'prop_malfunction': 0.0, # Percentage of defective agents
@@ -149,7 +148,7 @@ def main(argv):
# Build agent specific observations and normalize
for a in range(env.get_num_agents()):
- agent_next_obs[a] = normalize_observation(next_obs[a], observation_radius=10)
+ agent_next_obs[a] = normalize_observation(next_obs[a], tree_depth, observation_radius=10)
cummulated_reward[a] += all_rewards[a]
# Update replay buffer and train agent
@@ -186,7 +185,7 @@ def main(argv):
for _idx in range(env.get_num_agents()):
if done[_idx] == 1:
tasks_finished += 1
- done_window.append(tasks_finished / env.get_num_agents())
+ done_window.append(tasks_finished / max(1, env.get_num_agents()))
scores_window.append(score / max_steps) # save most recent score
scores.append(np.mean(scores_window))
dones_list.append((np.mean(done_window)))
diff --git a/utils/observation_utils.py b/utils/observation_utils.py
index e9eb3ed..ddb0374 100644
--- a/utils/observation_utils.py
+++ b/utils/observation_utils.py
@@ -89,7 +89,7 @@ def _split_subtree_into_feature_groups(node: TreeObsForRailEnv.Node, current_tre
if not node.childs:
return data, distance, agent_data
- for direction in TreeObsForRailEnv.tree_explorted_actions_char:
+ for direction in TreeObsForRailEnv.tree_explored_actions_char:
sub_data, sub_distance, sub_agent_data = _split_subtree_into_feature_groups(node.childs[direction], current_tree_depth + 1, max_tree_depth)
data = np.concatenate((data, sub_data))
distance = np.concatenate((distance, sub_distance))
--
GitLab