Skip to content
Snippets Groups Projects
Commit 13f392e4 authored by u214892's avatar u214892
Browse files

Merge branch 'master' of gitlab.aicrowd.com:flatland/flatland into 42-run-baselines-in-ci

parents 506aff90 3729c9d2
No related branches found
No related tags found
No related merge requests found
......@@ -30,10 +30,10 @@ Gives a global observation of the entire rail environment.
The observation is composed of the following elements:
- transition map array with dimensions (``env.height``, ``env.width``, ``16``), assuming **16 bits encoding of transitions**.
- Two 2D arrays (``map_height``, ``map_width``, ``2``) containing respectively the position of the given agent target and the positions of the other agents targets.
- A 3D array (``map_height``, ``map_width``, ``8``) with the **4 first channels** containing the **one hot encoding** of the direction of the given agent and the 4 second channels containing the positions of the other agents at their position coordinates.
- Two 2D arrays (``map_height``, ``map_width``, ``2``) containing respectively the position of the given agent target and the positions of the other agents' targets.
- A 3D array (``map_height``, ``map_width``, ``8``) with the **first 4 channels** containing the **one hot encoding** of the direction of the given agent and the second 4 channels containing the positions of the other agents at their position coordinates.
Feel free to enhance this observation with any layer you think might help solve the problem.
We encourage you to enhance this observation with any layer you think might help solve the problem.
It would also be possible to construct a global observation for a super agent that controls all agents at once.
Local Grid Observation
......@@ -51,20 +51,20 @@ We encourage you to come up with creative ways to overcome this problem. In the
Tree Observation
----------------
The tree observations is build by exploiting the graph structure of the railway network. The observation is generated by spanning a **4 branched tree** from the current position of the agent. Each branch follows the allowed transitions (backward branch only allowed at dead-ends) untill a cell with multiple allowed transitions is reached. Here the information gathered along the branch is stored as a node in the tree.
Figure bellow illustrates how such a tree observation is build:
The tree observation is built by exploiting the graph structure of the railway network. The observation is generated by spanning a **4 branched tree** from the current position of the agent. Each branch follows the allowed transitions (backward branch only allowed at dead-ends) until a cell with multiple allowed transitions is reached. Here the information gathered along the branch is stored as a node in the tree.
The figure below illustrates how the tree observation is built:
1. From Agent location probe all 4 directions (``L:Blue``, ``F:Green``, ``R:Purple``, ``B:Red``) starting with left and start branches when transition is allowed.
1. For each branch walk along the allowed transition till you reach a dead-end, switch or the target destination.
2. Create a node an fill in node information as stated below.
3. If max depth of tree is not reached and there are possible transistions start new branches and repeat above steps.
2. Fill up all non existing branches with -infinity such that tree size is invariant to number of possible transitions at branching points.
1. For each branch walk along the allowed transition until you reach a dead-end, switch or the target destination.
2. Create a node and fill in the node information as stated below.
3. If max depth of tree is not reached and there are possible transitions, start new branches and repeat the steps above.
2. Fill up all non existing branches with -infinity such that tree size is invariant to the number of possible transitions at branching points.
Note that we always start with the left branch according to the agent orientation. Thus the tree observation is independent of the orientation of cells and only consideres relative orientation of transition object to the agent.
Note that we always start with the left branch according to the agent orientation. Thus the tree observation is independent of the NESW orientation of cells, and only considers the transitions relative to the agent's orientation.
The colors in the figure bellow illustrate what branch the cell belongs two. If there are multiple colors in a cell, this cell is visited by different branches of the tree observation.
The right side of the figure shows the resulting tree of the railway network on the left. Cross means no branch was build. If a node has no children it either was a terminal node (dead-end, max depth reached or no transition possible). Circle indicate nodes that are filled with the corresponding infromation as stated below in Node Information.
The colors in the figure bellow illustrate what branch the cell belongs to. If there are multiple colors in a cell, this cell is visited by different branches of the tree observation.
The right side of the figure shows the resulting tree of the railway network on the left. Cross means no branch was built. If a node has no children it was either a terminal node (dead-end, max depth reached or no transition possible). A circle indicates a node filled with the corresponding information stated below in Node Information.
.. image:: https://i.imgur.com/sGBBhzJ.png
......@@ -77,20 +77,20 @@ Node Information
Each node is filled with information gathered along the path to the node. Currently each node contains 9 features:
- 1: if own target lies on the explored branch the current distance from the agent in number of cells is stored.
- 2: if another agents target is detected the distance in number of cells from current agent position is stored.
- 3: if another agent is detected the distance in number of cells from current agent position is stored.
- 2: if another agent's target is detected, the distance in number of cells from the current agent position is stored.
- 3: if another agent is detected, the distance in number of cells from the current agent position is stored.
- 4: possible conflict detected (This only works when we use a predictor and will not be important in this tutorial)
- 5: if an not usable switch (for agent) is detected we store the distance. An unusable switch is a switch where the agent does not have any choice of path, but other agents coming from different directions might.
- 5: if an unusable switch (for the agent) is detected we store the distance. An unusable switch is a switch where the agent does not have any choice of path, but other agents coming from different directions might.
- 6: This feature stores the distance (in number of cells) to the next node (e.g. switch or target or dead-end)
- 7: minimum remaining travel distance from node to the agent's target given the direction of the agent if this path is chosen
- 7: minimum remaining travel distance from this node to the agent's target given the direction of the agent if this path is chosen
- 8: agent in the same direction found on path to node
- ``n`` = number of agents present same direction (possible future use: number of other agents in the same direction in this branch)
- ``0`` = no agent present same direction
- ``n`` = number of agents present in the same direction (possible future use: number of other agents in the same direction in this branch)
- ``0`` = no agent present in the same direction
- 9: agent in the opposite direction on path to node
- ``n`` = number of agents present other direction than myself
- ``0`` = no agent present other direction than myself
- ``n`` = number of agents present in the opposite direction to the observing agent
- ``0`` = no agent present in other direction to the observing agent
from typing import Tuple
import numpy as np
from flatland.core.grid.grid4 import Grid4Transitions
from flatland.core.transition_map import GridTransitionMap
def make_simple_rail():
def make_simple_rail() -> Tuple[GridTransitionMap, np.array]:
# We instantiate a very simple rail network on a 7x10 grid:
# |
# |
......
......@@ -53,4 +53,4 @@ def test_walker():
print(obs_builder.distance_map[(0, *[0, 1], 1)])
assert obs_builder.distance_map[(0, *[0, 1], 1)] == 3
print(obs_builder.distance_map[(0, *[0, 2], 3)])
assert obs_builder.distance_map[(0, *[0, 2], 1)] == 2 # does not work yet, Erik's proposal.
assert obs_builder.distance_map[(0, *[0, 2], 1)] == 2
......@@ -3,9 +3,13 @@
import numpy as np
from flatland.core.grid.grid4 import Grid4TransitionsEnum
from flatland.envs.agent_utils import EnvAgent
from flatland.envs.generators import rail_from_GridTransitionMap_generator
from flatland.envs.observations import GlobalObsForRailEnv
from flatland.envs.rail_env import RailEnv
from flatland.envs.observations import GlobalObsForRailEnv, TreeObsForRailEnv
from flatland.envs.predictions import ShortestPathPredictorForRailEnv
from flatland.envs.rail_env import RailEnv, RailEnvActions
from flatland.utils.rendertools import RenderTool
from tests.simple_rail import make_simple_rail
"""Tests for `flatland` package."""
......@@ -35,3 +39,255 @@ def test_global_obs():
# If this assertion is wrong, it means that the observation returned
# places the agent on an empty cell
assert (np.sum(rail_map * global_obs[0][1][:, :, :4].sum(2)) > 0)
def _step_along_shortest_path(env, obs_builder, rail):
actions = {}
expected_next_position = {}
for agent in env.agents:
agent: EnvAgent
shortest_distance = np.inf
for exit_direction in range(4):
neighbour = obs_builder._new_position(agent.position, exit_direction)
if neighbour[0] >= 0 and neighbour[0] < env.height and neighbour[1] >= 0 and neighbour[1] < env.width:
desired_movement_from_new_cell = (exit_direction + 2) % 4
# Check all possible transitions in new_cell
for agent_orientation in range(4):
# Is a transition along movement `entry_direction' to the neighbour possible?
is_valid = obs_builder.env.rail.get_transition((neighbour[0], neighbour[1], agent_orientation),
desired_movement_from_new_cell)
if is_valid:
distance_to_target = obs_builder.distance_map[
(agent.handle, *agent.position, exit_direction)]
print("agent {} at {} facing {} taking {} distance {}".format(agent.handle, agent.position,
agent.direction,
exit_direction,
distance_to_target))
if distance_to_target < shortest_distance:
shortest_distance = distance_to_target
actions_to_be_taken_when_facing_north = {
Grid4TransitionsEnum.NORTH: RailEnvActions.MOVE_FORWARD,
Grid4TransitionsEnum.EAST: RailEnvActions.MOVE_RIGHT,
Grid4TransitionsEnum.WEST: RailEnvActions.MOVE_LEFT,
Grid4TransitionsEnum.SOUTH: RailEnvActions.DO_NOTHING,
}
print(" improved (direction) -> {}".format(exit_direction))
actions[agent.handle] = actions_to_be_taken_when_facing_north[
(exit_direction - agent.direction) % len(rail.transitions.get_direction_enum())]
expected_next_position[agent.handle] = neighbour
print(" improved (action) -> {}".format(actions[agent.handle]))
_, rewards, dones, _ = env.step(actions)
return rewards
def test_reward_function_conflict(rendering=False):
rail, rail_map = make_simple_rail()
env = RailEnv(width=rail_map.shape[1],
height=rail_map.shape[0],
rail_generator=rail_from_GridTransitionMap_generator(rail),
number_of_agents=2,
obs_builder_object=TreeObsForRailEnv(max_depth=2, predictor=ShortestPathPredictorForRailEnv()),
)
obs_builder: TreeObsForRailEnv = env.obs_builder
# initialize agents_static
env.reset()
# set the initial position
agent = env.agents_static[0]
agent.position = (5, 6) # south dead-end
agent.direction = 0 # north
agent.target = (3, 9) # east dead-end
agent.moving = True
agent = env.agents_static[1]
agent.position = (3, 8) # east dead-end
agent.direction = 3 # west
agent.target = (6, 6) # south dead-end
agent.moving = True
# reset to set agents from agents_static
env.reset(False, False)
if rendering:
renderer = RenderTool(env, gl="PILSVG")
renderer.renderEnv(show=True, show_observations=True)
iteration = 0
expected_positions = {
0: {
0: (5, 6),
1: (3, 8)
},
# both can move
1: {
0: (4, 6),
1: (3, 7)
},
# first can move, second stuck
2: {
0: (3, 6),
1: (3, 7)
},
# both stuck from now on
3: {
0: (3, 6),
1: (3, 7)
},
4: {
0: (3, 6),
1: (3, 7)
},
5: {
0: (3, 6),
1: (3, 7)
},
}
while iteration < 5:
rewards = _step_along_shortest_path(env, obs_builder, rail)
for agent in env.agents:
assert rewards[agent.handle] == -1
expected_position = expected_positions[iteration + 1][agent.handle]
assert agent.position == expected_position, "[{}] agent {} at {}, expected {}".format(iteration + 1,
agent.handle,
agent.position,
expected_position)
if rendering:
renderer.renderEnv(show=True, show_observations=True)
iteration += 1
def test_reward_function_waiting(rendering=False):
rail, rail_map = make_simple_rail()
env = RailEnv(width=rail_map.shape[1],
height=rail_map.shape[0],
rail_generator=rail_from_GridTransitionMap_generator(rail),
number_of_agents=2,
obs_builder_object=TreeObsForRailEnv(max_depth=2, predictor=ShortestPathPredictorForRailEnv()),
)
obs_builder: TreeObsForRailEnv = env.obs_builder
# initialize agents_static
env.reset()
# set the initial position
agent = env.agents_static[0]
agent.position = (3, 8) # east dead-end
agent.direction = 3 # west
agent.target = (3, 1) # west dead-end
agent.moving = True
agent = env.agents_static[1]
agent.position = (5, 6) # south dead-end
agent.direction = 0 # north
agent.target = (3, 8) # east dead-end
agent.moving = True
# reset to set agents from agents_static
env.reset(False, False)
if rendering:
renderer = RenderTool(env, gl="PILSVG")
renderer.renderEnv(show=True, show_observations=True)
iteration = 0
expectations = {
0: {
'positions': {
0: (3, 8),
1: (5, 6),
},
'rewards': [-1, -1],
},
1: {
'positions': {
0: (3, 7),
1: (4, 6),
},
'rewards': [-1, -1],
},
# second agent has to wait for first, first can continue
2: {
'positions': {
0: (3, 6),
1: (4, 6),
},
'rewards': [-1, -1],
},
# both can move again
3: {
'positions': {
0: (3, 5),
1: (3, 6),
},
'rewards': [-1, -1],
},
4: {
'positions': {
0: (3, 4),
1: (3, 7),
},
'rewards': [-1, -1],
},
# second reached target
5: {
'positions': {
0: (3, 3),
1: (3, 8),
},
'rewards': [-1, 0],
},
6: {
'positions': {
0: (3, 2),
1: (3, 8),
},
'rewards': [-1, 0],
},
# first reaches, target too
7: {
'positions': {
0: (3, 1),
1: (3, 8),
},
'rewards': [1, 1],
},
8: {
'positions': {
0: (3, 1),
1: (3, 8),
},
'rewards': [1, 1],
},
}
while iteration < 7:
rewards = _step_along_shortest_path(env, obs_builder, rail)
if rendering:
renderer.renderEnv(show=True, show_observations=True)
print(env.dones["__all__"])
for agent in env.agents:
agent: EnvAgent
print("[{}] agent {} at {}, target {} ".format(iteration + 1, agent.handle, agent.position, agent.target))
print(np.all([np.array_equal(agent2.position, agent2.target) for agent2 in env.agents]))
for agent in env.agents:
expected_position = expectations[iteration + 1]['positions'][agent.handle]
assert agent.position == expected_position, \
"[{}] agent {} at {}, expected {}".format(iteration + 1,
agent.handle,
agent.position,
expected_position)
expected_reward = expectations[iteration + 1]['rewards'][agent.handle]
actual_reward = rewards[agent.handle]
assert expected_reward == actual_reward, "[{}] agent {} reward {}, expected {}".format(iteration + 1,
agent.handle,
actual_reward,
expected_reward)
iteration += 1
0% Loading or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment