From 3579e545184c0a6a24e25065ba5cb86624579f75 Mon Sep 17 00:00:00 2001
From: MLErik <baerenjesus@gmail.com>
Date: Thu, 10 Oct 2019 10:55:38 -0400
Subject: [PATCH] fixed infeasible agents to always have feasible tasks
---
examples/introduction_flatland_2_1.py | 20 ++++-----
flatland/envs/schedule_generators.py | 42 +++++++++++--------
...est_flatland_envs_sparse_rail_generator.py | 2 +-
tests/test_flatland_malfunction.py | 4 +-
4 files changed, 38 insertions(+), 30 deletions(-)
diff --git a/examples/introduction_flatland_2_1.py b/examples/introduction_flatland_2_1.py
index 1e80021c..42917fb7 100644
--- a/examples/introduction_flatland_2_1.py
+++ b/examples/introduction_flatland_2_1.py
@@ -1,8 +1,6 @@
# In Flatland you can use custom observation builders and predicitors
# Observation builders generate the observation needed by the controller
# Preditctors can be used to do short time prediction which can help in avoiding conflicts in the network
-import time
-
from flatland.envs.observations import GlobalObsForRailEnv
# First of all we import the Flatland rail environment
from flatland.envs.rail_env import RailEnv
@@ -91,6 +89,7 @@ env_renderer = RenderTool(env, gl="PILSVG",
screen_width=1000) # Adjust these parameters to fit your resolution
+# The first thing we notice is that some agents don't have feasible paths to their target.
# We first look at the map we have created
# nv_renderer.render_env(show=True)
@@ -217,19 +216,19 @@ for agent_idx, agent in enumerate(env.agents):
for a in range(env.get_num_agents()):
action = controller.act(0)
action_dict.update({a: action})
-
# Do the environment step
observations, rewards, dones, information = env.step(action_dict)
-print("\n Thefollowing agents can register an action:")
+print("\n The following agents can register an action:")
print("========================================")
-print(information['action_required'])
+for info in information['action_required']:
+ print("Agent {} needs to submit an action.".format(info))
# We recommend that you monitor the malfunction data and the action required in order to optimize your training
# and controlling code.
# Let us now look at an episode playing out with random actions performed
-print("Start episode...")
+print("\nStart episode...")
# Reset the rendering system
env_renderer.reset()
@@ -237,10 +236,12 @@ env_renderer.reset()
# Here you can also further enhance the provided observation by means of normalization
# See training navigation example in the baseline repository
+
score = 0
# Run episode
frame_step = 0
-for step in range(500):
+
+for step in range(10):
# Chose an action for each agent in the environment
for a in range(env.get_num_agents()):
action = controller.act(observations[a])
@@ -248,10 +249,9 @@ for step in range(500):
# Environment step which returns the observations for all agents, their corresponding
# reward and whether their are done
- start_time = time.time()
+
next_obs, all_rewards, done, _ = env.step(action_dict)
- end_time = time.time()
- print(end_time - start_time)
+
# env_renderer.render_env(show=True, show_observations=False, show_predictions=False)
frame_step += 1
# Update replay buffer and train agent
diff --git a/flatland/envs/schedule_generators.py b/flatland/envs/schedule_generators.py
index ef05733d..ef27ac1a 100644
--- a/flatland/envs/schedule_generators.py
+++ b/flatland/envs/schedule_generators.py
@@ -82,29 +82,37 @@ def sparse_schedule_generator(speed_ratio_map: Mapping[float, float] = None, see
agents_position = []
agents_target = []
agents_direction = []
- for agent_idx in range(num_agents):
-
- # Set target for agent
- start_city = agent_start_targets_cities[agent_idx][0]
- target_city = agent_start_targets_cities[agent_idx][1]
- start_idx = np.random.choice(np.arange(len(train_stations[start_city])))
- target_idx = np.random.choice(np.arange(len(train_stations[target_city])))
- start = train_stations[start_city][start_idx]
- target = train_stations[target_city][target_idx]
+ for agent_idx in range(num_agents):
+ infeasible_agent = True
+ tries = 0
+ while infeasible_agent:
+ tries += 1
+ infeasible_agent = False
+ # Set target for agent
+ city_idx = np.random.randint(len(agent_start_targets_cities))
+ start_city = agent_start_targets_cities[city_idx][0]
+ target_city = agent_start_targets_cities[city_idx][1]
- while start[1] % 2 != 0:
start_idx = np.random.choice(np.arange(len(train_stations[start_city])))
- start = train_stations[start_city][start_idx]
- while target[1] % 2 != 1:
target_idx = np.random.choice(np.arange(len(train_stations[target_city])))
+ start = train_stations[start_city][start_idx]
target = train_stations[target_city][target_idx]
- agent_orientation = (agent_start_targets_cities[agent_idx][2] + 2 * start[1]) % 4
- if not rail.check_path_exists(start[0], agent_orientation, target[0]):
- agent_orientation = (agent_orientation + 2) % 4
- if not (rail.check_path_exists(start[0], agent_orientation, target[0])):
- warnings.warn("Infeasible task for agent {}".format(agent_idx))
+ while start[1] % 2 != 0:
+ start_idx = np.random.choice(np.arange(len(train_stations[start_city])))
+ start = train_stations[start_city][start_idx]
+ while target[1] % 2 != 1:
+ target_idx = np.random.choice(np.arange(len(train_stations[target_city])))
+ target = train_stations[target_city][target_idx]
+ agent_orientation = (agent_start_targets_cities[city_idx][2] + 2 * start[1]) % 4
+ if not rail.check_path_exists(start[0], agent_orientation, target[0]):
+ agent_orientation = (agent_orientation + 2) % 4
+ if not (rail.check_path_exists(start[0], agent_orientation, target[0])):
+ infeasible_agent = True
+ if tries >= 100:
+ warnings.warn("Did not find any possible path, check your parameters!!!")
+ break
agents_position.append((start[0][0], start[0][1]))
agents_target.append((target[0][0], target[0][1]))
diff --git a/tests/test_flatland_envs_sparse_rail_generator.py b/tests/test_flatland_envs_sparse_rail_generator.py
index 7bd6334b..f254e091 100644
--- a/tests/test_flatland_envs_sparse_rail_generator.py
+++ b/tests/test_flatland_envs_sparse_rail_generator.py
@@ -504,7 +504,7 @@ def test_sparse_rail_generator():
for a in range(env.get_num_agents()):
s0 = Vec2d.get_manhattan_distance(env.agents[a].initial_position, (0, 0))
s1 = Vec2d.get_chebyshev_distance(env.agents[a].initial_position, (0, 0))
- assert s0 == 33, "actual={}".format(s0)
+ assert s0 == 31, "actual={}".format(s0)
assert s1 == 24, "actual={}".format(s1)
diff --git a/tests/test_flatland_malfunction.py b/tests/test_flatland_malfunction.py
index e9b5a15d..8bbfefbb 100644
--- a/tests/test_flatland_malfunction.py
+++ b/tests/test_flatland_malfunction.py
@@ -122,11 +122,11 @@ def test_malfunction_process():
total_down_time += env.agents[0].malfunction_data['malfunction']
# Check that the appropriate number of malfunctions is achieved
- assert env.agents[0].malfunction_data['nr_malfunctions'] == 21, "Actual {}".format(
+ assert env.agents[0].malfunction_data['nr_malfunctions'] == 20, "Actual {}".format(
env.agents[0].malfunction_data['nr_malfunctions'])
# Check that 20 stops where performed
- assert agent_halts == 21
+ assert agent_halts == 20
# Check that malfunctioning data was standing around
assert total_down_time > 0
--
GitLab