#67 multiple agents

15a725f0 · u214892 · e959427a · 15a725f0 · 15a725f0
Commit 15a725f0 authored 5 years ago by u214892
--- a/notebooks/simple_example_3_manual_control.ipynb
+++ b/notebooks/simple_example_3_manual_control.ipynb
@@ -114,7 +114,7 @@
    "# Print the observation vector for agent 0\n",
    "obs, all_rewards, done, _ = env.step({0: 0})\n",
    "for i in range(env.get_num_agents()):\n",
-    "    env.obs_builder.util_print_obs_subtree(tree=obs[i], num_features_per_node=7)\n",
+    "    env.obs_builder.util_print_obs_subtree(tree=obs[i])\n",
    "\n",
    "env_renderer = RenderTool(env, gl=\"PIL\")\n",
    "# env_renderer = RenderTool(env, gl=\"PILSVG\")\n",

 %% Cell type:markdown id: tags:

 ### Simple Example 3 - Manual Control

 By default this runs a few "move forward" actions for two agents, in a separate window.

 If you uncomment the "input" line below, it opens a text box in the Jupyter notebook, allowing basic manual control.

 eg Enter `"0 2 s<enter>"` to tell agent 0 to move forward, and step the environment.

 You should be able to see the red agent step forward, and get a reward from the env, looking like this:

 `Rewards:  {0: -1.0, 1: -1.0}   [done= {0: False, 1: False, '__all__': False} ]`

 Note that this example is set up to use the straightforward "PIL" renderer - without the special SBB artwork!
 The agent observations are displayed as squares of varying sizes, with a paler version of the agent colour.  The targets are half-size squares in the full agent colour.

 You can switch to the "PILSVG" renderer which is prettier but currently renders the agents one step behind, because it needs to know which way the agent is turning.  This can be confusing if you are debugging step-by-step.

 The image below is what the separate window should look like.

 %% Cell type:markdown id: tags:

 ![simple_example_3.png](simple_example_3.png)

 %% Cell type:code id: tags:

 ``` python
 import random
 import numpy as np
 import time
 from flatland.envs.generators import random_rail_generator
 from flatland.envs.observations import TreeObsForRailEnv
 from flatland.envs.rail_env import RailEnv
 from flatland.utils.rendertools import RenderTool
 ```

 %% Cell type:code id: tags:

 ``` python
 random.seed(1)
 np.random.seed(1)

 env = RailEnv(width=7,
              height=7,
              rail_generator=random_rail_generator(),
              number_of_agents=2,
              obs_builder_object=TreeObsForRailEnv(max_depth=2))

 # Print the observation vector for agent 0
 obs, all_rewards, done, _ = env.step({0: 0})
 for i in range(env.get_num_agents()):
-    env.obs_builder.util_print_obs_subtree(tree=obs[i], num_features_per_node=7)
+    env.obs_builder.util_print_obs_subtree(tree=obs[i])

 env_renderer = RenderTool(env, gl="PIL")
 # env_renderer = RenderTool(env, gl="PILSVG")

 env_renderer.renderEnv(show=True, frames=True)

 print("Manual control: s=perform step, q=quit, [agent id] [1-2-3 action] \
       (turnleft+move, move to front, turnright+move)")
 ```

 %% Output

     [0, 0, 0, 0, 3.0, 0, 0]
      L: [0, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
      F: [-inf, inf, inf, inf, 2, 1.0, 0]
        L: [0, 0, 1, inf, inf, 3, 0]
        F: [0, 0, inf, inf, inf, 5, 8.0]
        R: [0, 0, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
      R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
      B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
     [0, 0, 0, 0, 8.0, 0, 0]
      L: [0, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
      F: [-inf, inf, inf, inf, 2, 6.0, 0]
        L: [0, 0, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, inf, inf, inf, 3, 5.0]
        R: [0, 0, inf, inf, inf, 6, 6.0]
        B: [0, 0, -inf, -inf, -inf, -inf, -inf]
      R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
      B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        L: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        F: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        R: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
        B: [-inf, -inf, -inf, -inf, -inf, -inf, -inf]
    Manual control: s=perform step, q=quit, [agent id] [1-2-3 action]        (turnleft+move, move to front, turnright+move)

 %% Cell type:code id: tags:

 ``` python
 for step in range(10):

    # This is an example command, setting agent 0's action to 2 (move forward), and agent 1's action to 2,
    # then stepping the environment.
    cmd = "0 2 1 2 s"

    # uncomment this input statement if you want to try interactive manual commands
    # cmd = input(">> ")

    cmds = cmd.split(" ")

    action_dict = {}

    i = 0
    while i < len(cmds):
        if cmds[i] == 'q':
            import sys

            sys.exit()
        elif cmds[i] == 's':
            obs, all_rewards, done, _ = env.step(action_dict)
            action_dict = {}
            print("Rewards: ", all_rewards, "  [done=", done, "]")
        else:
            agent_id = int(cmds[i])
            action = int(cmds[i + 1])
            action_dict[agent_id] = action
            i = i + 1
        i += 1

    env_renderer.renderEnv(show=True, frames=True)

    time.sleep(0.3)
 ```

 %% Output

    Rewards:  {0: -1.0, 1: -1.0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: -1.0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: -1.0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: -1.0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: 0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: 0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]
    Rewards:  {0: 0, 1: 0}   [done= {0: False, 1: False, '__all__': False} ]

--- a/tox.ini
+++ b/tox.ini
@@ -113,8 +113,6 @@ deps =
    -r{toxinidir}/requirements_dev.txt
    -r{toxinidir}/requirements_continuous_integration.txt
 commands =
-; install current version of flatland to be used by notebooks
-    sh -c 'python setup.py install'
 ; run tests from subfolder to ensure that resources are accessed via resources and not via relative paths
    sh -c 'mkdir -p {envtmpdir}/6f59bc68108c3895b1828abdd04b9a06'
    sh -c 'jupyter nbextension install --py --sys-prefix widgetsnbextension'