CI test for RL examples: Drive and VehicleFollowing

.github/workflows/ci-base-tests-linux.yml

@@ -6,7 +6,7 @@ env:
   venv_dir: .venv
 
 jobs:
-  base-tests-linux:
+  base-tests:
     runs-on: ubuntu-20.04
     if: github.event_name == 'push' || github.event.pull_request.head.repo.full_name != github.repository
     container: ghcr.io/smarts-project/smarts:v0.6.1-minimal
@@ -19,8 +19,8 @@ jobs:
           - ./smarts/env --ignore=./smarts/env/tests/test_rllib_hiway_env.py
           - ./smarts/env/tests/test_rllib_hiway_env.py
           - ./smarts/sstudio
-          - ./examples/tests --ignore=./examples/tests/test_learning.py
           - ./smarts/ray
+          - ./examples/tests/test_examples.py
     steps:
       - name: Checkout
         uses: actions/checkout@v2
@@ -31,12 +31,25 @@ jobs:
           pip install --upgrade pip
           pip install wheel==0.38.4
           pip install -e .[camera_obs,opendrive,test,test_notebook,torch,train,gym,argoverse,envision,sumo]
-          if echo ${{matrix.tests}} | grep -q -e "env" -e "examples"; then pip install -e .[rllib]; fi
-          if echo ${{matrix.tests}} | grep -q "/ray"; then pip install -e .[ray]; fi
+          if echo ${{matrix.tests}} | grep -q -e "test_rllib_hiway_env.py" -e "test_examples.py"; then pip install -e .[rllib]; fi
+          if echo ${{matrix.tests}} | grep -q -e "/smarts/ray"; then pip install -e .[ray]; fi
+      - name: Build scenarios
+        run: |
+          . ${{env.venv_dir}}/bin/activate
+          scl scenario build-all \
+            scenarios/open_drive/od_4lane \
+            scenarios/open_drive/od_merge \
+            scenarios/sumo/figure_eight \
+            scenarios/sumo/intersections/2lane \
+            scenarios/sumo/intersections/4lane \
+            scenarios/sumo/intersections/6lane \
+            scenarios/sumo/loop \
+            scenarios/sumo/straight/3lane_bubble \
+            scenarios/sumo/tests/multi_agents_loop \
+            scenarios/sumo/zoo_intersection
       - name: Run smoke tests
         run: |
           . ${{env.venv_dir}}/bin/activate
-          make build-all-scenarios
           PYTHONPATH=$PWD PYTHONHASHSEED=42 pytest -v \
             --doctest-modules \
             --forked \
@@ -49,5 +62,36 @@ jobs:
             --ignore=./smarts/core/tests/test_smarts_memory_growth.py \
             --ignore=./smarts/core/tests/test_env_frame_rate.py \
             --ignore=./smarts/env/tests/test_benchmark.py \
-            --ignore=./examples/tests/test_learning.py \
             -k 'not test_long_determinism'
+
+  examples-rl:
+    runs-on: ubuntu-20.04
+    if: github.event_name == 'push' || github.event.pull_request.head.repo.full_name != github.repository
+    container: ghcr.io/smarts-project/smarts:v0.6.1-minimal
+    strategy:
+      matrix:
+        tests:
+          - drive
+          - platoon
+    steps:
+      - name: Checkout
+        uses: actions/checkout@v2
+      - name: Install dependencies
+        run: |
+          cd ${GITHUB_WORKSPACE}/examples/rl/${{matrix.tests}}
+          python3.8 -m venv ${{env.venv_dir}}
+          . ${{env.venv_dir}}/bin/activate
+          pip install --upgrade pip
+          pip install wheel==0.38.4
+          pip install -e ./../../../.[camera_obs,argoverse,sumo,test]
+          pip install -e ./inference/    
+      - name: Run smoke tests
+        run: |
+          cd ${GITHUB_WORKSPACE}/examples/rl/${{matrix.tests}}
+          . ${{env.venv_dir}}/bin/activate
+          PYTHONPATH=$PWD PYTHONHASHSEED=42 pytest -v \
+            --doctest-modules \
+            --forked \
+            --dist=no \
+            -n auto \
+            ${GITHUB_WORKSPACE}/examples/tests/test_rl.py::test_${{matrix.tests}}

cli/tests/test_studio.py

@@ -53,10 +53,10 @@ def test_scenario_generation_unchanged():
             _hashseed = os.getenv("PYTHONHASHSEED")
             assert _hashseed not in (None, "random"), f"PYTHONHASHSEED is {_hashseed}"
 
-            shutil.copytree("scenarios/sumo", loc1)
+            shutil.copytree("scenarios/sumo/loop", loc1)
             build_scenarios(scenarios=[loc1], clean=True, seed=42)
 
-            shutil.copytree("scenarios/sumo", loc2)
+            shutil.copytree("scenarios/sumo/loop", loc2)
             build_scenarios(scenarios=[loc2], clean=True, seed=42)
 
             for dirpath, dirnames, files in os.walk(loc1):

envision/tests/test_data_formatter.py

@@ -45,7 +45,6 @@
 from smarts.core.scenario import Scenario
 from smarts.core.smarts import SMARTS
 from smarts.core.sumo_traffic_simulation import SumoTrafficSimulation
-from smarts.core.tests.helpers.bubbles import bubble_geometry
 from smarts.core.tests.helpers.scenario import temp_scenario
 from smarts.core.utils.file import unpack
 from smarts.sstudio.genscenario import gen_scenario
@@ -321,7 +320,7 @@ def bubble():
         actor=t.BoidAgentActor(
             # TODO: Provide a more self-contained way to build agent locators for tests
             name="hive-mind",
-            agent_locator="scenarios.sumo.straight.agent_prefabs:pose-boid-agent-v0",
+            agent_locator="scenarios.sumo.straight.3lane_bubble.agent_prefabs:pose-boid-agent-v0",
         ),
     )
 

examples/tests/test_rl.py

@@ -0,0 +1,59 @@
+import argparse
+from pathlib import Path
+from unittest import mock
+
+from smarts.core.utils import import_utils
+
+import_utils.import_module_from_file(
+    "examples", Path(__file__).parents[1] / "__init__.py"
+)
+
+
+def _mock_load_config(load_config):
+    def func():
+        config = load_config()
+        config.alg["n_steps"] = 4
+        config.alg["batch_size"] = 2
+        config.alg["n_epochs"] = 2
+        config.epochs = 2 * len(config.scenarios)
+        config.train_steps = 8
+        config.checkpoint_freq = 10000
+        config.eval_freq = 10000
+        return config
+
+    return func
+
+
+def test_platoon():
+    """Tests RL training of `examples/rl/platoon` example."""
+
+    from examples.rl.platoon.train.run import load_config, main
+
+    args = argparse.Namespace()
+    args.mode = "train"
+    args.logdir = None
+    args.model = None
+    args.head = False
+
+    with mock.patch(
+        "examples.rl.platoon.train.run.load_config",
+        _mock_load_config(load_config),
+    ):
+        main(args)
+
+
+def test_drive():
+    """Tests RL training of `examples/rl/drive` example."""
+    from examples.rl.drive.train.run import load_config, main
+
+    args = argparse.Namespace()
+    args.mode = "train"
+    args.logdir = None
+    args.model = None
+    args.head = False
+
+    with mock.patch(
+        "examples.rl.drive.train.run.load_config",
+        _mock_load_config(load_config),
+    ):
+        main(args)

examples/tools/pybullet_linear_tire_example.py

@@ -1,6 +1,5 @@
 import math
 from pathlib import Path
-from unittest import mock
 
 import matplotlib.pyplot as plt
 import numpy as np

scenarios/sumo/intersections/1_to_3lane_left_turn_middle_lane_c_agents_1/scenario.py

@@ -64,7 +64,7 @@
     ("E8", 0, "E4", 0),
 ]
 
-# Total route combinations = 14C1 + 14C2 + 14C3 + 14C4 = 1470
+# Total route combinations
 all_routes = vertical_routes + horizontal_routes + turn_left_routes + turn_right_routes
 route_comb = [com for elems in range(4, 5) for com in combinations(all_routes, elems)]
 traffic = {}

scenarios/sumo/intersections/1_to_3lane_left_turn_sumo_c_agents_1/scenario.py

@@ -64,7 +64,7 @@
     ("E8", 0, "E4", 0),
 ]
 
-# Total route combinations = 14C1 + 14C2 + 14C3 + 14C4 = 1470
+# Total route combinations
 all_routes = vertical_routes + horizontal_routes + turn_left_routes + turn_right_routes
 route_comb = [com for elems in range(4, 5) for com in combinations(all_routes, elems)]
 traffic = {}

scenarios/sumo/straight/cruise_2lane_agents_1/scenario.py

@@ -45,8 +45,7 @@
     (1, 1),
 ]
 
-# Traffic combinations = 3C2 + 3C3 = 3 + 1 = 4
-# Repeated traffic combinations = 4 * 100 = 400
+# Traffic combinations
 min_flows = 2
 max_flows = 2
 route_comb = [

scenarios/sumo/straight/cutin_2lane_agents_1/scenario.py

@@ -33,8 +33,7 @@
     (1, 1),
 ]
 
-# Traffic combinations = 3C2 + 3C3 = 3 + 1 = 4
-# Repeated traffic combinations = 4 * 100 = 400
+# Traffic combinations
 min_flows = 1
 max_flows = 3
 route_comb = [

scenarios/sumo/straight/overtake_2lane_agents_1/scenario.py

@@ -45,8 +45,7 @@
     (1, 1),
 ]
 
-# Traffic combinations = 3C2 + 3C3 = 3 + 1 = 4
-# Repeated traffic combinations = 4 * 100 = 400
+# Traffic combinations
 min_flows = 2
 max_flows = 2
 route_comb = [

scenarios/sumo/vehicle_following/merge_exit_sumo_agents_1/scenario.py

@@ -30,34 +30,14 @@
     speed=Distribution(sigma=0.2, mean=0.8),
 )
 
-# Social path = (start_lane, end_lane)
-social_paths = [
-    (0, 0),
-    (0, 1),
-    (0, 2),
-    (1, 0),
-    (1, 1),
-    (1, 2),
-    (2, 0),
-    (2, 1),
-    (2, 2),
-]
-min_flows = 2
-max_flows = 4
-social_comb = [
-    com
-    for elems in range(min_flows, max_flows + 1)
-    for com in combinations(social_paths, elems)
-]
-
 # Leader path = (start_lane, end_lane)
 leader_paths = [0, 1, 2]
 
 # Overall routes
-route_comb = product(social_comb, leader_paths)
+route_comb = leader_paths
 
 traffic = {}
-for name, (social_path, leader_path) in enumerate(route_comb):
+for name, leader_path in enumerate(route_comb):
     traffic[str(name)] = Traffic(
         engine="SUMO",
         flows=[],

scenarios/sumo/vehicle_following/straight_2lane_sumo_agents_1/scenario.py

@@ -25,29 +25,14 @@
 
 leader = TrafficActor(name="Leader-007", depart_speed=0)
 
-# Social path = (start_lane, end_lane)
-social_paths = [
-    (0, 0),
-    (0, 1),
-    (1, 0),
-    (1, 1),
-]
-min_flows = 2
-max_flows = 3
-social_comb = [
-    com
-    for elems in range(min_flows, max_flows + 1)
-    for com in combinations(social_paths, elems)
-] * 10
-
 # Leader path = (start_lane, end_lane)
 leader_paths = [(0, 0), (0, 1)]
 
 # Overall routes
-route_comb = product(social_comb, leader_paths)
+route_comb = leader_paths
 
 traffic = {}
-for name, (social_path, leader_path) in enumerate(route_comb):
+for name, leader_path in enumerate(route_comb):
     traffic[str(name)] = Traffic(
         engine="SUMO",
         flows=[],

scenarios/sumo/vehicle_following/straight_3lanes_sumo_agents_1/scenario.py

@@ -26,23 +26,13 @@
 
 leader = TrafficActor(name="Leader-007", depart_speed=0)
 
-# Social path = (start_lane, end_lane)
-social_paths = [(0, 0), (0, 2), (2, 0), (2, 2)]
-min_flows = 2
-max_flows = 4
-social_comb = [
-    com
-    for elems in range(min_flows, max_flows + 1)
-    for com in combinations(social_paths, elems)
-]
-
 # Leader path = (start_lane, end_lane)
 leader_paths = [(0, 0), (0, 1), (0, 2), (1, 0), (1, 1), (1, 2), (2, 0), (2, 1), (2, 2)]
 
 # Overall routes
-route_comb = product(social_comb, leader_paths)
+route_comb = leader_paths
 traffic = {}
-for name, (social_path, leader_path) in enumerate(route_comb):
+for name, leader_path in enumerate(route_comb):
     traffic[str(name)] = Traffic(
         engine="SUMO",
         flows=[],

smarts/core/tests/test_bubble_manager.py

@@ -27,14 +27,10 @@
 from helpers.scenario import temp_scenario
 
 import smarts.sstudio.types as t
-from smarts.core.agent_manager import AgentManager
 from smarts.core.controllers import ActionSpaceType
 from smarts.core.coordinates import Heading, Pose
-from smarts.core.local_traffic_provider import LocalTrafficProvider
-from smarts.core.observations import Observation
 from smarts.core.scenario import Scenario
 from smarts.core.smarts import SMARTS
-from smarts.core.sumo_traffic_simulation import SumoTrafficSimulation
 from smarts.core.tests.helpers.providers import MockProvider, MockTrafficProvider
 from smarts.core.vehicle_index import VehicleIndex
 from smarts.sstudio import gen_scenario

zoo/evaluation/scenarios/cross/ego_agent.py

@@ -1,5 +1,4 @@
 from smarts.core.agent_interface import ActionSpaceType, AgentInterface
-from smarts.core.utils.episodes import episodes
 from smarts.zoo.agent_spec import AgentSpec
 from zoo.policies.non_interactive_agent import NonInteractiveAgent