Merge branch 'master' of github.com:ucd-dare/CarDreamer

Author: gaodechen

README.md

@@ -107,6 +107,8 @@ The following table shows the overall performance metrics over different CarDrea
 
 CarDreamer enables the customization of different levels of observability. The table below highlights performance metrics under different observability settings, including full observability, field-of-view (FOV), and recursive field-of-view (SFOV). These settings allow agents to operate with varying degrees of environmental awareness, impacting their ability to plan and execute maneuvers effectively.
 
+Check out [common.yaml](https://github.com/ucd-dare/CarDreamer/blob/master/car_dreamer/configs/common.yaml) for different birdeye observability and entity settings. The entity settings allow users to determine whether to render, for example, background vehicle intentions (planned trajectories), ego vechile intentions (planned trajectories), and transmission errors for intention sharing.
+
 | Observability Performance Metrics |
 | :-------------------------------: |
 | ![Observability Performance](https://ucd-dare.github.io/cardreamer.github.io/static/images/tables/observability_performance.png) |

car_dreamer/configs/common.yaml

@@ -3,7 +3,7 @@ env:
 
   world:
     carla_port: 2000
-    town: "Town04" # which town to simulate
+    town: "Town04" # which town map to use
     actor_active_distance: 100
     fixed_delta_seconds: 0.1
     auto_lane_change: True
@@ -163,7 +163,10 @@ env:
     image_size: 512
 
   action:
-    discrete: True # whether to use discrete control space
+    discrete: False
+    # Discrete control space. Use continuos_acc and continuous_steer if discrete is False
     discrete_acc: [-2.0, 0.0, 2.0] # discrete value of accelerations
     discrete_steer: [-0.6, -0.2, 0.0, 0.2, 0.6] # discrete value of steering angles
-    n_commands: 3 # number of text commands
+    continuous_acc: [-3.0, 3.0] # continuous value of accelerations
+    continuous_steer: [-1.0, 1.0] # continuous value of steering angles
+    n_commands: 3 # number of text commands for message handler

car_dreamer/configs/tasks.yaml

@@ -379,6 +379,7 @@ carla_stop_sign:
     decoder.cnn_keys: "birdeye_with_traffic_lights"
     decoder.cnn_kernels: [5, 5, 5, 6, 6]
     train.log_keys_video: [camera, birdeye_with_traffic_lights]
+
 carla_follow:
   env:
     world:
@@ -403,7 +404,6 @@ carla_follow:
         distance: 2.0
       }
 
-
     terminal:
       time_limit: 500 # maximum timesteps per episode
       out_lane_thres: 5 # threshold for out of lane

car_dreamer/toolkit/observer/handlers/renderer/constants.py

@@ -74,13 +74,13 @@ class Color:
 
 
 class BirdeyeEntity(Enum):
-    ROADMAP = "roadmap"
+    ROADMAP = "roadmap"  # Roadmap obtained from CARLA map topology
     EGO_VEHICLE = "ego_vehicle"
     BACKGROUND_VEHICLES = "background_vehicles"
-    FOV_LINES = "fov_lines"
-    WAYPOINTS = "waypoints"
-    BACKGROUND_WAYPOINTS = "background_waypoints"
+    FOV_LINES = "fov_lines"  # Field of view lines that indicate the visibility of the ego vehicle
+    WAYPOINTS = "waypoints"  # Waypoints that indicate the path of the ego vehicle
+    BACKGROUND_WAYPOINTS = "background_waypoints"  # Waypoints (intentions) of the background vehicles
     TRAFFIC_LIGHTS = "traffic_lights"
     STOP_SIGNS = "stop_signs"
     MESSAGES = "messages"
-    ERROR_BACKGROUND_WAYPOINTS = "error_background_waypoints"
+    ERROR_BACKGROUND_WAYPOINTS = "error_background_waypoints"  # Waypoints that may have transmission errors

docs/source/tasks.rst

@@ -101,6 +101,8 @@ Here is the descriptions of some common configurations in ``car_dreamer/configs/
       If ``True``, ``len(discrete_acc) * len(discrete_steer)`` is the size of the action space.
     * ``discrete_acc`` (default: ``[-2.0, 0.0, 2.0]``): the discrete values for acceleration.
     * ``discrete_steer`` (default: ``[-0.6, -0.2, 0.0, 0.2, 0.6]``): the discrete values for steering.
+    * ``continuous_acc`` (default: ``[-3.0, 3.0]``): the continuous value range for acceleration.
+    * ``continuous_steer`` (default: ``[-1.0, 1.0]``): the continuous value range for steering.
 
 Tasks and Environments
 --------------------------

dreamerv3/README.md

@@ -8,7 +8,7 @@ First, install the required dependencies for DreamerV3:
 cd dreamerv3
 conda install -c "nvidia/label/cuda-11.8.0" cuda-toolkit
 pip install -r requirements.txt
-pip install --upgrade "jax[cuda12_pip]" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
+pip install "jax[cuda12_pip]==0.4.34" -f https://storage.googleapis.com/jax-releases/jax_cuda_releases.html
 ```
 
 Set up CARLA and environment variables:

dreamerv3/eval_stats.py

@@ -123,7 +123,7 @@ def main(args):
         ],
         [
             "Avg. Destination Reached",
-            f"{final_metrics['avg_destination_reached']:.2f} ± {final_metrics['avg_destination_reached_sem']:.2f}",
+            f"{final_metrics['avg_destination_reached']:.2%} ± {final_metrics['avg_destination_reached_sem']:.2%}",
         ],
         [
             "Avg. Time Exceeded",

dreamerv3/requirements.txt

@@ -2,8 +2,8 @@ cloudpickle==3.0.0
 crafter==1.8.3
 gym==0.12.5
 dm_control==1.0.18
-jax==0.4.26
-jaxlib==0.4.26
+jax==0.4.34
+jaxlib==0.4.34
 numpy==1.24.3
 optax==0.2.2
 rich==13.7.1
@@ -20,3 +20,6 @@ shapely==1.8.4
 carla==0.9.15
 matplotlib==3.8.4
 tabulate==0.9.0
+msgpack
+zmq
+einops