From ccd2d08bcf51db081c60edd52bb9a3b94f015ed3 Mon Sep 17 00:00:00 2001
From: Erik Nygren <erik.nygren@sbb.ch>
Date: Sat, 5 Oct 2019 11:31:40 -0400
Subject: [PATCH] minor bugfixes
---
torch_training/training_navigation.py | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/torch_training/training_navigation.py b/torch_training/training_navigation.py
index 4f82c52..ac80feb 100644
--- a/torch_training/training_navigation.py
+++ b/torch_training/training_navigation.py
@@ -115,9 +115,6 @@ def main(argv):
# Reset environment
obs, info = env.reset(True, True)
- register_action_state = np.zeros(env.get_num_agents(), dtype=bool)
- final_obs = agent_obs.copy()
- final_obs_next = agent_next_obs.copy()
# Build agent specific observations
for a in range(env.get_num_agents()):
@@ -155,8 +152,11 @@ def main(argv):
# Update replay buffer and train agent
for a in range(env.get_num_agents()):
- if (agent_obs_buffer[a] is not None and register_action_state[a] and env.agents[a].status != 3) or \
+ if (agent_obs_buffer[a] is not None and info['action_required'][a] and env.agents[a].status != 3) or \
env.agents[a].status == 2:
+ if all_rewards[a] < -1.:
+ print("bad")
+
agent_delayed_next = agent_obs[a].copy()
agent.step(agent_obs_buffer[a], agent_action_buffer[a], all_rewards[a],
agent_delayed_next, done[a])
--
GitLab