Update devkit v0.3

.env

@@ -0,0 +1,10 @@
+# Common variables
+NUPLAN_TEST_MAPS_VERSION="nuplan-maps-v1.0"
+
+# Simulation variables
+SIMULATION_CONTAINER_PORT=50050
+SIMULATION_IMAGE="nuplan/nuplan"
+
+# Submission variables
+SUBMISSION_CONTAINER_PORT=50051
+SUBMISSION_IMAGE="test-contestant/nuplan-test-submission"

Dockerfile.submission

@@ -0,0 +1,49 @@
+FROM ubuntu:20.04
+
+RUN apt-get update \
+    && apt-get install -y curl gnupg2 software-properties-common default-jdk
+
+ENV APT_KEY_DONT_WARN_ON_DANGEROUS_USAGE=DontWarn
+
+RUN curl -fsSL https://bazel.build/bazel-release.pub.gpg | apt-key add - \
+    && curl -fsSL https://nvidia.github.io/nvidia-docker/gpgkey | apt-key add - \
+    && curl -s -L https://nvidia.github.io/nvidia-docker/ubuntu20.04/nvidia-docker.list | tee /etc/apt/sources.list.d/nvidia-docker.list \
+    && add-apt-repository "deb [arch=amd64] https://storage.googleapis.com/bazel-apt stable jdk1.8" \
+    && apt-get update \
+    && apt-get install -y \
+        bazel \
+        file \
+        zip \
+        nvidia-container-toolkit \
+        software-properties-common \
+    && rm -rf /var/lib/apt/lists/*
+
+# Download miniconda and install silently.
+ENV PATH /opt/conda/bin:$PATH
+RUN curl -fsSLo Miniconda3-latest-Linux-x86_64.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh && \
+    bash Miniconda3-latest-Linux-x86_64.sh -b -p /opt/conda && \
+    rm Miniconda3-latest-Linux-x86_64.sh && \
+    conda clean -a -y
+
+# Setup working environement
+ARG NUPLAN_HOME=/nuplan_devkit
+WORKDIR $NUPLAN_HOME
+
+COPY requirements.txt requirements_submission.txt $NUPLAN_HOME/
+
+RUN bash -c "python -m pip install --upgrade pip --pre && \
+    pip install --no-cache-dir -r $NUPLAN_HOME/requirements.txt -f https://download.pytorch.org/whl/torch_stable.html && \
+    pip install --no-cache-dir -r $NUPLAN_HOME/requirements_submission.txt"
+
+RUN mkdir -p $NUPLAN_HOME/nuplan
+
+COPY setup.py $NUPLAN_HOME
+COPY nuplan $NUPLAN_HOME/nuplan
+
+RUN bash -c "pip install -e ."
+
+ENV NUPLAN_MAPS_ROOT=/data/sets/nuplan/maps \
+    NUPLAN_DATA_ROOT=/data/sets/nuplan \
+    NUPLAN_EXP_ROOT=/data/exp/nuplan
+
+CMD ["/bin/bash", "-c", "python nuplan/submission/submission_planner.py"]

Jenkinsfile

@@ -9,7 +9,7 @@ pipeline{
     kubernetes(
       jnlp.nuplan_devkit(
         name: 'nuplan-devkit-tests',
-        tag: "v1.0.2",
+        tag: "v1.0.3",
         cpu: 8, maxcpu: 8,
         memory: "32G", maxmemory: "64G", yaml: """spec:
   containers:

README.md

@@ -32,6 +32,8 @@ Stay tuned for the upcoming nuPlan ML planning challenges!
 ______________________________________________________________________
 
 ## Changelog
+- Jul 15 2022
+  * v0.3 Devkit: Updates to the devit including but not limited to: nuBoard update, reduce RAM usage during caching and training, new map APIs, simulation and metrics runtime improvements. 
 - Jun 10 2022
   * v1.0 Dataset: Full nuPlan dataset release with over 1,300 hours of driving data (15,000+ logs) across 4 cities (Las Vegas, Pittsburgh, Boston, Singapore).
   * v0.2 Devkit: Multiple devkit updates with improvements across the whole framework (planning models, training, simulation, metrics, dashboard, tutorials and more).

docker-compose.yml

@@ -0,0 +1,33 @@
+version: "3.7"
+
+x-nuplan-volumes: &x-nuplan-volumes
+  volumes:
+    - "$NUPLAN_DATA_ROOT:/data/sets/nuplan:rw"
+
+
+services:
+  submission:
+    <<: *x-nuplan-volumes
+    container_name: "test-contestant_nuplan-test-submission"
+    image: ${SUBMISSION_IMAGE}
+    env_file:
+      - .env
+    ports:
+      - ${SUBMISSION_CONTAINER_PORT}:${SUBMISSION_CONTAINER_PORT}
+
+    tty: true
+
+  simulation:
+    <<: *x-nuplan-volumes
+    env_file:
+      - .env
+    environment:
+      'NUPLAN_CHALLENGE': $NUPLAN_CHALLENGE
+      'NUPLAN_PLANNER': $NUPLAN_PLANNER
+    image: ${SIMULATION_IMAGE}
+    ports:
+      - ${SIMULATION_CONTAINER_PORT}:${SIMULATION_CONTAINER_PORT}
+    tty: true
+    depends_on:
+      - submission
+    entrypoint: /nuplan_devkit/nuplan/entrypoint.sh

docs/dataset_setup.md

@@ -32,7 +32,7 @@ The final hierarchy should look as follows (depending on the splits downloaded a
     │   └── us-pa-pittsburgh-hazelwood
     │       └── 9.17.1937
     │           └── map.gpkg
-    └── nuplan-v1.0
+    └── nuplan_v1.0
         ├── mini
         │   ├── 2021.05.12.22.00.38_veh-35_01008_01518.db
         │   ├── 2021.06.09.17.23.18_veh-38_00773_01140.db

docs/installation.md

@@ -38,9 +38,9 @@ Next we setup a virtual environment. We recommend Conda for this purpose.
 See the [official Miniconda page](https://conda.io/en/latest/miniconda.html).
 
 ### Create a Conda environment
-We create a new Conda environment named `nuplan`.
+We create a new Conda environment using environment.yml.
 ```
-conda create --name nuplan python=3.9
+conda env create -f environment.yml
 ```
 
 ### Activate the environment

environment.yml

@@ -0,0 +1,10 @@
+name: nuplan
+channels:
+  - conda-forge
+dependencies:
+  - python=3.9
+  - pip=21.2.4
+  - nb_conda_kernels
+  - pip:
+    - -r requirements_torch.txt
+    - -r requirements.txt

nuplan/common/actor_state/BUILD

@@ -84,6 +84,7 @@ py_library(
     name = "agent",
     srcs = ["agent.py"],
     deps = [
+        ":agent_temporal_state",
         ":oriented_box",
         ":scene_object",
         ":state_representation",
@@ -128,3 +129,25 @@ py_library(
         "//nuplan/common/utils:split_state",
     ],
 )
+
+py_library(
+    name = "agent_temporal_state",
+    srcs = ["agent_temporal_state.py"],
+    deps = [
+        "//nuplan/common/actor_state:oriented_box",
+        "//nuplan/common/actor_state:scene_object",
+        "//nuplan/common/actor_state:tracked_objects_types",
+        "//nuplan/planning/simulation/trajectory:predicted_trajectory",
+    ],
+)
+
+py_library(
+    name = "ego_temporal_state",
+    srcs = ["ego_temporal_state.py"],
+    deps = [
+        "//nuplan/common/actor_state:agent_temporal_state",
+        "//nuplan/common/actor_state:ego_state",
+        "//nuplan/common/actor_state:tracked_objects_types",
+        "//nuplan/planning/simulation/trajectory:predicted_trajectory",
+    ],
+)

nuplan/common/actor_state/agent.py

@@ -3,14 +3,15 @@
 import copy
 from typing import List, Optional
 
+from nuplan.common.actor_state.agent_temporal_state import AgentTemporalState
 from nuplan.common.actor_state.oriented_box import OrientedBox
 from nuplan.common.actor_state.scene_object import SceneObject, SceneObjectMetadata
-from nuplan.common.actor_state.state_representation import StateSE2, StateVector2D
+from nuplan.common.actor_state.state_representation import StateSE2, StateVector2D, TimePoint
 from nuplan.common.actor_state.tracked_objects_types import TrackedObjectType
 from nuplan.planning.simulation.trajectory.predicted_trajectory import PredictedTrajectory
 
 
-class Agent(SceneObject):
+class Agent(AgentTemporalState, SceneObject):
     """Class describing Agents in the scene, representing Vehicles, Bicycles and Pedestrians"""
 
     def __init__(
@@ -32,14 +33,18 @@ def __init__(
         :param angular_velocity: The scalar angular velocity of the agent, if available
         :param predictions: Optional list of (possibly multiple) predicted trajectories
         """
-        super().__init__(tracked_object_type, oriented_box, metadata)
+        AgentTemporalState.__init__(
+            self,
+            initial_time_stamp=TimePoint(metadata.timestamp_us),
+            predictions=predictions,
+            past_trajectory=past_trajectory,
+        )
+        SceneObject.__init__(
+            self, tracked_object_type=tracked_object_type, oriented_box=oriented_box, metadata=metadata
+        )
         self._velocity = velocity
         self._angular_velocity = angular_velocity
 
-        # Possible multiple predicted trajectories
-        self._predictions: List[PredictedTrajectory] = predictions if predictions is not None else []
-        self._past_trajectory = past_trajectory
-
     @property
     def velocity(self) -> StateVector2D:
         """
@@ -56,57 +61,6 @@ def angular_velocity(self) -> Optional[float]:
         """
         return self._angular_velocity
 
-    @property
-    def predictions(self) -> List[PredictedTrajectory]:
-        """
-        Getter for agents predicted trajectories
-        :return: Trajectories
-        """
-        return self._predictions
-
-    @predictions.setter
-    def predictions(self, predicted_trajectories: List[PredictedTrajectory]) -> None:
-        """
-        Setter for predicted trajectories, checks if the listed probabilities sum to one.
-        :param predicted_trajectories: List of Predicted trajectories
-        """
-        # Sanity check that if predictions are provided, probabilities sum to 1
-        probability_sum = sum(prediction.probability for prediction in predicted_trajectories)
-        if not abs(probability_sum - 1) < 1e-6 and predicted_trajectories:
-            raise ValueError(f"The provided trajectory probabilities did not sum to one, but to {probability_sum:.2f}!")
-        self._predictions = predicted_trajectories
-
-    @property
-    def past_trajectory(self) -> Optional[PredictedTrajectory]:
-        """
-        Getter for agents predicted trajectories
-        :return: Trajectories
-        """
-        return self._past_trajectory
-
-    @past_trajectory.setter
-    def past_trajectory(self, past_trajectory: Optional[PredictedTrajectory]) -> None:
-        """
-        Setter for predicted trajectories, checks if the listed probabilities sum to one.
-        :param past_trajectory: Driven Trajectory
-        """
-        if not past_trajectory:
-            # In case it is none, no check is needed
-            self._past_trajectory = past_trajectory
-            return
-
-        # Make sure that the current state is set!
-        last_waypoint = past_trajectory.waypoints[-1]
-        if not last_waypoint:
-            raise RuntimeError(
-                f"Last waypoint represents current agent's {self.metadata.track_id} state, this should not be None!"
-            )
-
-        # Sanity check that last waypoint is at the same time index as the current one
-        if last_waypoint.time_us != self.metadata.timestamp_us:
-            raise ValueError("The provided trajectory does not end at current agent state!")
-        self._past_trajectory = past_trajectory
-
     @classmethod
     def from_new_pose(cls, agent: Agent, pose: StateSE2) -> Agent:
         """

nuplan/common/actor_state/agent_temporal_state.py

@@ -0,0 +1,86 @@
+from __future__ import annotations
+
+from typing import List, Optional
+
+from nuplan.common.actor_state.state_representation import TimePoint
+from nuplan.planning.simulation.trajectory.predicted_trajectory import PredictedTrajectory
+
+
+class AgentTemporalState:
+    """
+    Actor with current, multimodal future as well as past trajectory.
+        The future trajectory probabilities have to sum up to 1.0.
+        The past trajectory is only single modal with mode probability 1.0.
+        The last waypoint in past trajectory has to be the same as current position (we check only timestamp).
+    """
+
+    def __init__(
+        self,
+        initial_time_stamp: TimePoint,
+        predictions: Optional[List[PredictedTrajectory]] = None,
+        past_trajectory: Optional[PredictedTrajectory] = None,
+    ):
+        """
+        Initialize actor temporal state which has past as well as future trajectory
+        :param initial_time_stamp: time stamp the current detections
+        :param predictions: future multimodal trajectory
+        :param past_trajectory: past trajectory transversed
+        """
+        self._initial_time_stamp = initial_time_stamp
+        self.predictions: List[PredictedTrajectory] = predictions if predictions is not None else []
+        self.past_trajectory = past_trajectory
+
+    @property
+    def predictions(self) -> List[PredictedTrajectory]:
+        """
+        Getter for agents predicted trajectories
+        :return: Trajectories
+        """
+        return self._predictions
+
+    @predictions.setter
+    def predictions(self, predicted_trajectories: List[PredictedTrajectory]) -> None:
+        """
+        Setter for predicted trajectories, checks if the listed probabilities sum to one.
+        :param predicted_trajectories: List of Predicted trajectories
+        """
+        if not predicted_trajectories:
+            self._predictions = predicted_trajectories
+            return
+        # Sanity check that if predictions are provided, probabilities sum to 1
+        probability_sum = sum(prediction.probability for prediction in predicted_trajectories)
+        if not abs(probability_sum - 1) < 1e-6 and predicted_trajectories:
+            raise ValueError(f"The provided trajectory probabilities did not sum to one, but to {probability_sum:.2f}!")
+        self._predictions = predicted_trajectories
+
+    @property
+    def past_trajectory(self) -> Optional[PredictedTrajectory]:
+        """
+        Getter for agents predicted trajectories
+        :return: Trajectories
+        """
+        return self._past_trajectory
+
+    @past_trajectory.setter
+    def past_trajectory(self, past_trajectory: Optional[PredictedTrajectory]) -> None:
+        """
+        Setter for predicted trajectories, checks if the listed probabilities sum to one.
+        :param past_trajectory: Driven Trajectory
+        """
+        if not past_trajectory:
+            # In case it is none, no check is needed
+            self._past_trajectory = past_trajectory
+            return
+
+        # Make sure that the current state is set!
+        last_waypoint = past_trajectory.waypoints[-1]
+        if not last_waypoint:
+            raise RuntimeError("Last waypoint represents current agent's state, this should not be None!")
+
+        # Sanity check that last waypoint is at the same time index as the current one
+        if last_waypoint.time_point != self._initial_time_stamp:
+            raise ValueError(
+                "The provided trajectory does not end at current agent state!"
+                f" {last_waypoint.time_us} != {self._initial_time_stamp}"
+            )
+        self._past_trajectory = past_trajectory