Release 0 6 1 rc1 (huawei-noah#1396)

Co-authored-by: Allan Liu <[email protected]> Co-authored-by: qianyi-sun <[email protected]> Co-authored-by: adai <[email protected]> Co-authored-by: Saul Field <[email protected]> Co-authored-by: Rutvik Gupta <[email protected]>
idsc-frazzoli · Apr 19, 2022 · 3ef5650 · 3ef5650
1 parent 7ff9f0a
commit 3ef5650
Show file tree

Hide file tree

Showing 21 changed files with 107 additions and 98 deletions.
diff --git a/.github/workflows/ci-format.yml b/.github/workflows/ci-format.yml
@@ -1,12 +1,6 @@
 name: SMARTS CI Format
 
-on:
-  push:
-    branches-ignore:
-      - master
-  pull_request:
-    branches-ignore:
-      - master
+on: [push, pull_request]
 
 env:
   venv_dir: .venv

diff --git a/CHANGELOG.md b/CHANGELOG.md
@@ -10,6 +10,13 @@ Copy and pasting the git commit messages is __NOT__ enough.
 
 # [Unreleased]
 
+### [0.6.1rc1] 15-04-18
+### Fixed
+- Gracefully handle `EndlessGoal` missions in the MaRL benchmark. Relative goal distance with `EndlessGoal` will be now always be 0.
+- Restore `rl-agent` to working order. Version incremented to `1.1.0`.
+- Update `rl-agent` wheel.
+- Add workaround for issue where traffic history starts offroad by fixing `scl scenario build`'s `--allow-offset-map` to work with traffic history.
+
 ### [0.6.1rc0] 15-04-16
 ### Added
 - Added `get_vehicle_start_time()` method for scenarios with traffic history data.  See Issue #1210.

diff --git a/baselines/marl_benchmark/common.py b/baselines/marl_benchmark/common.py
@@ -171,10 +171,11 @@ def cal_goal_relative_pos(env_obs: Observation, _):
 
         ego_state = env_obs.ego_vehicle_state
         goal = ego_state.mission.goal
-        assert isinstance(goal, PositionalGoal), goal
 
         ego_pos = ego_state.position[:2]
-        goal_pos = goal.position  # the position of mission goal is 2-dimensional.
+        goal_pos = getattr(
+            goal, "position", ego_pos
+        )  # the position of mission goal is 2-dimensional.
         vector = np.asarray([goal_pos[0] - ego_pos[0], goal_pos[1] - ego_pos[1]])
         # space = SPACE_LIB["goal_relative_pos"](None)
         # return vector / (space.high - space.low)

diff --git a/baselines/marl_benchmark/metrics/basic_handler.py b/baselines/marl_benchmark/metrics/basic_handler.py
@@ -41,8 +41,8 @@ def agent_info_adapter(env_obs, shaped_reward: float, raw_info: dict):
     info["collision"] = 1 if len(env_obs.events.collisions) > 0 else 0
 
     goal = env_obs.ego_vehicle_state.mission.goal
-    goal_pos = goal.position
     ego_2d_pos = env_obs.ego_vehicle_state.position[:2]
+    goal_pos = getattr(goal, "position", ego_2d_pos)
 
     info["distance_to_goal"] = distance.euclidean(ego_2d_pos, goal_pos)
     info["distance_to_center"] = CalObs.cal_distance_to_center(env_obs, "")

diff --git a/baselines/marl_benchmark/wrappers/rllib/frame_stack.py b/baselines/marl_benchmark/wrappers/rllib/frame_stack.py
@@ -234,20 +234,14 @@ def func(env_obs_seq, env_reward):
             # ======== Penalty: distance to goal =========
             goal = last_env_obs.ego_vehicle_state.mission.goal
             ego_2d_position = last_env_obs.ego_vehicle_state.position[:2]
-            if hasattr(goal, "position"):
-                goal_position = goal.position
-            else:
-                goal_position = ego_2d_position
+            goal_position = getattr(goal, "position", ego_2d_position)
             goal_dist = distance.euclidean(ego_2d_position, goal_position)
             penalty += -0.01 * goal_dist
 
             old_obs = env_obs_seq[-2]
             old_goal = old_obs.ego_vehicle_state.mission.goal
             old_ego_2d_position = old_obs.ego_vehicle_state.position[:2]
-            if hasattr(old_goal, "position"):
-                old_goal_position = old_goal.position
-            else:
-                old_goal_position = old_ego_2d_position
+            old_goal_position = getattr(old_goal, "position", old_ego_2d_position)
             old_goal_dist = distance.euclidean(old_ego_2d_position, old_goal_position)
             penalty += 0.1 * (old_goal_dist - goal_dist)  # 0.05
 

diff --git a/cli/studio.py b/cli/studio.py
@@ -71,7 +71,7 @@ def _build_single_scenario(clean: bool, allow_offset_map: bool, scenario: str):
     from smarts.core.scenario import Scenario
 
     traffic_histories = Scenario.discover_traffic_histories(scenario_root_str)
-    shift_to_origin = not allow_offset_map or bool(traffic_histories)
+    shift_to_origin = not allow_offset_map
 
     map_spec = Scenario.discover_map(scenario_root_str, shift_to_origin=shift_to_origin)
     road_map, _ = map_spec.builder_fn(map_spec)

diff --git a/docs/sim/scenario_studio.rst b/docs/sim/scenario_studio.rst
@@ -24,7 +24,7 @@ Then in the scenario.py file:
 
     t.SocialAgentActor(
         name="my-rl-agent",
-        agent_locator="rl_agent:rl_agent-v0"
+        agent_locator="rl_agent:rl_agent-v1"
     )
 
 

diff --git a/examples/rllib/__init__.py b/examples/rllib/__init__.py
diff --git a/requirements.txt b/requirements.txt
@@ -59,7 +59,7 @@ hyperlink==21.0.0
 idna==3.3
 ijson==3.1.4
 importlib-metadata==4.11.3
-importlib-resources==5.7.0
+importlib-resources==5.7.1
 incremental==21.3.0
 iniconfig==1.1.1
 ipykernel==6.13.0
@@ -70,7 +70,7 @@ jsonpatch==1.32
 jsonpointer==2.3
 jsonschema==4.4.0
 jupyter-client==7.2.2
-jupyter-core==4.9.2
+jupyter-core==4.10.0
 jupyter-server==1.16.0
 jupyter-server-mathjax==0.2.5
 jupyterlab-pygments==0.2.2
@@ -170,7 +170,7 @@ tornado==6.1
 traitlets==5.1.1
 trimesh==3.9.29
 Twisted==22.4.0
-typing_extensions==4.1.1
+typing_extensions==4.2.0
 urllib3==1.26.9
 visdom==0.1.8.9
 wcwidth==0.2.5

diff --git a/scenarios/NGSIM/README.md b/scenarios/NGSIM/README.md
@@ -16,6 +16,10 @@ Once that is done, you should be able to build your scenarios in the normal way,
 ```bash
 scl scenario build-all --clean scenarios/NGSIM
 ```
+or the following command to disable generating `shifted_map-AUTOGEN.net.xml`
+```bash
+scl scenario build-all --clean --allow-offset-map scenarios/NGSIM
+```
 
 For each traffic history `.yml` file specified in your `scenario.py`, 
 this will creates a cooresponding `.shf` file that SMARTS will use

diff --git a/scenarios/NGSIM/us101/map.net.xml b/scenarios/NGSIM/us101/map.net.xml
@@ -1,20 +1,20 @@
 <?xml version="1.0" encoding="UTF-8"?>
 
-<!-- generated on 2022-03-11 18:22:26 by Eclipse SUMO netconvert Version 1.10.0
+<!-- generated on 2022-03-28 22:28:01 by Eclipse SUMO netedit Version 1.12.0
 <configuration xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="http://sumo.dlr.de/xsd/netconvertConfiguration.xsd">
 
     <input>
-        <sumo-net-file value="NGSIM/us101/map.net.xml"/>
+        <sumo-net-file value="scenarios/NGSIM/us101/map.net.xml"/>
     </input>
 
     <output>
-        <output-file value="NGSIM/us101/shifted_map-AUTOGEN.net.xml"/>
+        <output-file value="scenarios/NGSIM/us101/map.net.xml"/>
     </output>
 
     <processing>
         <geometry.min-radius.fix.railways value="false"/>
         <geometry.max-grade.fix value="false"/>
-        <offset.disable-normalization value="false"/>
+        <offset.disable-normalization value="true"/>
         <lefthand value="false"/>
     </processing>
 
@@ -39,34 +39,34 @@
 
 <net version="1.9" junctionCornerDetail="5" limitTurnSpeed="5.50" xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance" xsi:noNamespaceSchemaLocation="http://sumo.dlr.de/xsd/net_file.xsd">
 
-    <location netOffset="-17.37,-3.00" convBoundary="0.00,0.00,5.05,656.00" origBoundary="-10000000000.00,-10000000000.00,10000000000.00,10000000000.00" projParameter="!"/>
+    <location netOffset="-17.37,-3.00" convBoundary="17.12,0.00,22.55,656.00" origBoundary="-10000000000.00,-10000000000.00,10000000000.00,10000000000.00" projParameter="!"/>
 
     <edge id="E0" from="J0" to="J1" priority="-1">
-        <lane id="E0_0" index="0" speed="13.89" length="656.00" width="3.50" shape="-15.62,656.00 -15.62,0.00"/>
-        <lane id="E0_1" index="1" speed="13.89" length="656.00" width="3.50" shape="-12.12,656.00 -12.12,0.00"/>
-        <lane id="E0_2" index="2" speed="13.89" length="656.00" width="3.50" shape="-8.62,656.00 -8.62,0.00"/>
-        <lane id="E0_3" index="3" speed="13.89" length="656.00" width="3.50" shape="-5.12,656.00 -5.12,0.00"/>
-        <lane id="E0_4" index="4" speed="13.89" length="656.00" width="3.50" acceleration="1" shape="-1.62,656.00 -1.62,0.00"/>
+        <lane id="E0_0" index="0" speed="13.89" length="656.00" width="3.50" shape="1.88,656.00 1.88,0.00"/>
+        <lane id="E0_1" index="1" speed="13.89" length="656.00" width="3.50" shape="5.38,656.00 5.38,0.00"/>
+        <lane id="E0_2" index="2" speed="13.89" length="656.00" width="3.50" shape="8.88,656.00 8.88,0.00"/>
+        <lane id="E0_3" index="3" speed="13.89" length="656.00" width="3.50" shape="12.38,656.00 12.38,0.00"/>
+        <lane id="E0_4" index="4" speed="13.89" length="656.00" width="3.50" acceleration="1" shape="15.88,656.00 15.88,0.00"/>
     </edge>
-    <edge id="E2" from="J7" to="J8" priority="-1" shape="2.75,132.35 1.78,143.30 1.32,149.16 0.82,156.44 0.42,163.86 0.03,171.62 0.03,179.98 0.02,187.62 0.01,196.23 0.00,413.13">
-        <lane id="E2_0" index="0" speed="13.89" length="280.88" width="3.70" shape="4.59,132.51 3.62,143.46 3.17,149.29 2.67,156.55 2.27,163.96 1.88,171.67 1.88,179.98 1.87,187.62 1.86,196.23 1.85,411.65"/>
+    <edge id="E2" from="J7" to="J8" priority="-1" shape="20.25,132.35 19.32,141.71 18.86,146.52 18.26,153.27 17.43,160.48 17.23,171.73 17.12,178.90 17.22,185.68 17.21,195.58 17.50,413.13">
+        <lane id="E2_0" index="0" speed="13.89" length="280.93" width="3.70" shape="22.09,132.53 21.16,141.89 20.70,146.69 20.10,153.46 19.28,160.60 19.08,171.76 18.97,178.90 19.07,185.67 19.06,195.58 19.35,411.64"/>
     </edge>
     <edge id="E3" from="J9" to="J10" priority="-1">
-        <lane id="E3_0" index="0" speed="13.89" length="41.51" shape="6.64,132.56 1.69,173.77"/>
+        <lane id="E3_0" index="0" speed="13.89" length="41.51" shape="24.14,132.56 19.19,173.77"/>
     </edge>
     <edge id="E4" from="J8" to="J11" priority="-1">
-        <lane id="E4_0" index="0" speed="13.89" length="84.64" width="4.00" shape="2.07,414.51 6.38,497.56"/>
+        <lane id="E4_0" index="0" speed="13.89" length="84.64" width="4.00" shape="19.57,414.51 23.88,497.56"/>
     </edge>
 
-    <junction id="J0" type="dead_end" x="0.13" y="656.00" incLanes="" intLanes="" shape="0.13,656.00 -15.87,656.00" customShape="1"/>
-    <junction id="J1" type="dead_end" x="0.13" y="0.00" incLanes="E0_0 E0_1 E0_2 E0_3 E0_4" intLanes="" shape="-16.07,0.00 0.63,0.00" customShape="1"/>
-    <junction id="J10" type="dead_end" x="0.10" y="173.58" incLanes="E3_0" intLanes="" shape="3.28,173.96 0.10,173.58"/>
-    <junction id="J11" type="dead_end" x="4.38" y="497.66" incLanes="E4_0" intLanes="" shape="8.37,497.45 4.38,497.66"/>
-    <junction id="J7" type="dead_end" x="2.75" y="132.35" incLanes="" intLanes="" shape="2.75,132.35 6.44,132.68" fringe="inner"/>
-    <junction id="J8" type="priority" x="0.00" y="413.13" incLanes="E2_0" intLanes="" shape="0.08,414.61 4.07,414.40 3.70,411.65 0.00,411.65">
+    <junction id="J0" type="dead_end" x="17.63" y="656.00" incLanes="" intLanes="" shape="17.63,656.00 0.00,656.00" customShape="1"/>
+    <junction id="J1" type="dead_end" x="17.63" y="0.00" incLanes="E0_0 E0_1 E0_2 E0_3 E0_4" intLanes="" shape="0.00,0.00 18.13,0.00" customShape="1"/>
+    <junction id="J10" type="dead_end" x="17.60" y="173.58" incLanes="E3_0" intLanes="" shape="20.78,173.96 17.60,173.58"/>
+    <junction id="J11" type="dead_end" x="21.88" y="497.66" incLanes="E4_0" intLanes="" shape="25.87,497.45 21.88,497.66"/>
+    <junction id="J7" type="dead_end" x="20.25" y="132.35" incLanes="" intLanes="" shape="20.25,132.35 23.93,132.72" fringe="inner"/>
+    <junction id="J8" type="priority" x="17.50" y="413.13" incLanes="E2_0" intLanes="" shape="17.58,414.61 21.57,414.40 21.20,411.64 17.50,411.64">
         <request index="0" response="0" foes="0"/>
     </junction>
-    <junction id="J9" type="dead_end" x="5.05" y="132.37" incLanes="" intLanes="" shape="5.05,132.37 8.23,132.75"/>
+    <junction id="J9" type="dead_end" x="22.55" y="132.37" incLanes="" intLanes="" shape="22.55,132.37 25.73,132.75"/>
 
     <connection from="E2" to="E4" fromLane="0" toLane="0" dir="s" state="M"/>
 

diff --git a/setup.py b/setup.py
@@ -13,7 +13,7 @@
     description="Scalable Multi-Agent RL Training School",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    version="0.6.1rc0",
+    version="0.6.1rc1",
     packages=find_packages(exclude=("tests", "examples")),
     include_package_data=True,
     zip_safe=True,

diff --git a/smarts/core/utils/episodes.py b/smarts/core/utils/episodes.py
@@ -69,7 +69,7 @@ def record_step(self, observations=None, rewards=None, dones=None, infos=None):
         """Record a step end."""
         self.steps += 1
 
-        if not isinstance(observations, dict):
+        if not isinstance(dones, dict):
             observations, rewards, dones, infos = self._convert_to_dict(
                 observations, rewards, dones, infos
             )

diff --git a/smarts/ros/src/smarts_ros/package.xml b/smarts/ros/src/smarts_ros/package.xml
@@ -1,7 +1,7 @@
 <?xml version="1.0"?>
 <package format="2">
   <name>smarts_ros</name>
-  <version>0.6.1rc0</version>
+  <version>0.6.1rc1</version>
   <description>The SMARTS package includes a node to wrap and drive a SMARTS simulation,
     publishing the state updates after each step.</description>
 

diff --git a/zoo/evaluation/batch_list.yaml b/zoo/evaluation/batch_list.yaml
@@ -74,7 +74,7 @@ agent_list:
           rate: 1
 
   group_3:
-    - locator: rl_agent:rl-agent-v0
+    - locator: rl_agent:rl-agent-v1
       name: normal_speed_agent
       params:
         goal_is_nearby_threshold: 40
@@ -86,7 +86,7 @@ agent_list:
         target_speed: 15
         lane_change_speed: 12.5
 
-    - locator: rl_agent:rl-agent-v0
+    - locator: rl_agent:rl-agent-v1
       name: low_speed_agent
       params:
         goal_is_nearby_threshold: 40

diff --git a/.../rl-agent/rl_agent-1.0.0-py3-none-any.whl → .../rl-agent/rl_agent-1.1.0-py3-none-any.whl b/.../rl-agent/rl_agent-1.0.0-py3-none-any.whl → .../rl-agent/rl_agent-1.1.0-py3-none-any.whl
diff --git a/zoo/policies/rl-agent/rl_agent/__init__.py b/zoo/policies/rl-agent/rl_agent/__init__.py
@@ -2,9 +2,6 @@
 
 import rl_agent
 
-from smarts.zoo.agent_spec import AgentSpec
-from smarts.zoo.registry import register
-
 from .agent import RLAgent
 from .lane_space import (
     ACTION_SPACE,
@@ -14,41 +11,47 @@
     get_observation_adapter,
 )
 
-VERSION = "1.0.0"
-
-
-def entrypoint(
-    goal_is_nearby_threshold=40,
-    lane_end_threshold=51,
-    lane_crash_distance_threshold=6,
-    lane_crash_ttc_threshold=2,
-    intersection_crash_distance_threshold=6,
-    intersection_crash_ttc_threshold=5,
-    target_speed=15,
-    lane_change_speed=12.5,
-):
-    with pkg_resources.path(rl_agent, "checkpoint") as checkpoint_path:
-        return AgentSpec(
-            interface=agent_interface,
-            observation_adapter=get_observation_adapter(
-                goal_is_nearby_threshold=goal_is_nearby_threshold,
-                lane_end_threshold=lane_end_threshold,
-                lane_crash_distance_threshold=lane_crash_distance_threshold,
-                lane_crash_ttc_threshold=lane_crash_ttc_threshold,
-                intersection_crash_distance_threshold=intersection_crash_distance_threshold,
-                intersection_crash_ttc_threshold=intersection_crash_ttc_threshold,
-            ),
-            action_adapter=get_action_adapter(
-                target_speed=target_speed,
-                lane_change_speed=lane_change_speed,
-            ),
-            agent_builder=lambda: RLAgent(
-                load_path=str((checkpoint_path / "checkpoint").absolute()),
-                policy_name="default_policy",
-                observation_space=OBSERVATION_SPACE,
-                action_space=ACTION_SPACE,
-            ),
-        )
-
-
-register(locator="rl-agent-v1", entry_point=entrypoint)
+VERSION = "1.1.0"
+
+try:
+    from smarts.zoo.agent_spec import AgentSpec
+    from smarts.zoo.registry import register
+
+    def entrypoint(
+        goal_is_nearby_threshold=40,
+        lane_end_threshold=51,
+        lane_crash_distance_threshold=6,
+        lane_crash_ttc_threshold=2,
+        intersection_crash_distance_threshold=6,
+        intersection_crash_ttc_threshold=5,
+        target_speed=15,
+        lane_change_speed=12.5,
+    ):
+        with pkg_resources.path(rl_agent, "checkpoint") as checkpoint_path:
+            return AgentSpec(
+                interface=agent_interface,
+                observation_adapter=get_observation_adapter(
+                    goal_is_nearby_threshold=goal_is_nearby_threshold,
+                    lane_end_threshold=lane_end_threshold,
+                    lane_crash_distance_threshold=lane_crash_distance_threshold,
+                    lane_crash_ttc_threshold=lane_crash_ttc_threshold,
+                    intersection_crash_distance_threshold=intersection_crash_distance_threshold,
+                    intersection_crash_ttc_threshold=intersection_crash_ttc_threshold,
+                ),
+                action_adapter=get_action_adapter(
+                    target_speed=target_speed,
+                    lane_change_speed=lane_change_speed,
+                ),
+                agent_builder=lambda: RLAgent(
+                    load_path=str((checkpoint_path / "checkpoint").absolute()),
+                    policy_name="default_policy",
+                    observation_space=OBSERVATION_SPACE,
+                    action_space=ACTION_SPACE,
+                ),
+            )
+
+    register(locator="rl-agent-v1", entry_point=entrypoint)
+except:
+
+    def entrypoint(**args):
+        raise ImportError("SMARTS has not been installed. Use `pip install smarts`")
diff --git a/zoo/policies/rl-agent/rl_agent/agent.py b/zoo/policies/rl-agent/rl_agent/agent.py
@@ -5,7 +5,9 @@
 from pathlib import Path
 
 import gym
-import tensorflow as tf
+import tensorflow.compat.v1 as tf
+
+tf.disable_v2_behavior()
 from ray.rllib.agents.ppo.ppo_tf_policy import PPOTFPolicy as LoadPolicy
 from ray.rllib.models import ModelCatalog
 
@@ -31,13 +33,18 @@ def __init__(self, load_path, policy_name, observation_space, action_space):
             return
 
         self._prep = ModelCatalog.get_preprocessor_for_space(self._observation_space)
-        self._sess = tf.compat.v1.Session(graph=tf.Graph())
+        self._sess = tf.Session(graph=tf.Graph())
         self._sess.__enter__()
 
         with tf.name_scope(self._policy_name):
             # obs_space need to be flattened before passed to PPOTFPolicy
             flat_obs_space = self._prep.observation_space
-            self.policy = LoadPolicy(flat_obs_space, self._action_space, {})
+
+            self.policy = LoadPolicy(
+                flat_obs_space,
+                self._action_space,
+                {},
+            )
             objs = pickle.load(open(self._checkpoint_path, "rb"))
             objs = pickle.loads(objs["worker"])
             state = objs["state"]

diff --git a/zoo/policies/rl-agent/rl_agent/lane_space.py b/zoo/policies/rl-agent/rl_agent/lane_space.py
@@ -534,7 +534,7 @@ def observation_adapter(env_obs):
 
         # get goal
         goal = env_obs.ego_vehicle_state.mission.goal
-        goal_position = goal.position
+        goal_position = getattr(goal, "position", ego_state.position[:2])
 
         (
             lane_ttc,

diff --git a/zoo/policies/rl-agent/setup.py b/zoo/policies/rl-agent/setup.py
@@ -7,5 +7,5 @@
     version=VERSION,
     packages=["rl_agent"],
     include_package_data=True,
-    install_requires=["smarts", "tensorflow==2.2.1", "ray[rllib]==1.0.1.post1"],
+    install_requires=["smarts>=0.6.1rc1", "tensorflow==2.4", "ray[rllib]==1.0.1.post1"],
 )