updated versions in rosinstall | added seconds to model name | removed wandb because its missing in stable baselines branch

Author: ReykCS

.rosinstall

@@ -38,7 +38,7 @@
 - git:
     local-name: ../utils/arena-utils
     uri: https://github.com/Arena-Rosnav/arena-utils.git
-    version: v2.1.0
+    version: v2.2.0
 
 - git:
     local-name: ../utils/task-generator
@@ -55,7 +55,7 @@
 - git:
     local-name: ../planners/rosnav
     uri: https://github.com/Arena-Rosnav/rosnav.git
-    version: v1.1.1
+    version: v1.1.2
 
 - git:
     local-name: ../planners/arena-ros

arena_bringup/launch/start_arena.launch

@@ -9,14 +9,19 @@
   <arg name="task_id" default="" />
   <arg name="app_token" default="" />
   <arg name="app_token_key" default="" />
-  <arg name="task_finished_url" default="" />
+  <arg name="base_url" default="" />
+  <arg name="task_finished_endpoint" default="" />
+  <arg name="new_best_model_endpoint" default="" />
 
+  <param name="is_webapp_docker" value="$(arg is_webapp_docker)" />
   <param name="task_id" value="$(arg task_id)" />
   <param name="app_token" value="$(arg app_token)" />
   <param name="app_token_key" value="$(arg app_token_key)" />
-  <param name="task_finished_url" value="$(arg task_finished_url)" />
+  <param name="base_url" value="$(arg base_url)" />
+  <param name="task_finished_endpoint" value="$(arg task_finished_endpoint)" />
+  <param name="new_best_model_endpoint" value="$(arg new_best_model_endpoint)" />
 
-  <node name="task_progress_publisher" type="task_progress_publisher.py" pkg="task_progress_publisher" if="$(eval arg('is_webapp_docker') == true)" />
+  <node name="task_progress_publisher" type="task_progress_publisher.py" pkg="arena-utils" if="$(eval arg('is_webapp_docker') == true)" />
   <!-- -->
 
   <arg name="desired_resets" default="2" />

arena_bringup/launch/start_training.launch

@@ -5,12 +5,17 @@
   <arg name="task_id" default="" />
   <arg name="app_token" default="" />
   <arg name="app_token_key" default="" />
-  <arg name="task_finished_url" default="" />
+  <arg name="base_url" default="" />
+  <arg name="task_finished_endpoint" default="" />
+  <arg name="new_best_model_endpoint" default="" />
 
+  <param name="is_webapp_docker" value="$(arg is_webapp_docker)" />
   <param name="task_id" value="$(arg task_id)" />
   <param name="app_token" value="$(arg app_token)" />
   <param name="app_token_key" value="$(arg app_token_key)" />
-  <param name="task_finished_url" value="$(arg task_finished_url)" />
+  <param name="base_url" value="$(arg base_url)" />
+  <param name="task_finished_endpoint" value="$(arg task_finished_endpoint)" />
+  <param name="new_best_model_endpoint" value="$(arg new_best_model_endpoint)" />
 
   <node name="task_progress_publisher" type="task_progress_publisher.py" pkg="arena-utils" if="$(eval arg('is_webapp_docker') == true)" />
   <!-- -->

training/configs/training_config.yaml

@@ -10,7 +10,7 @@ no_gpu: false
 ### Training Monitoring
 monitoring:
   # weights and biases logging
-  use_wandb: true
+  use_wandb: false
   # save evaluation stats during training in log file
   eval_log: false
 
@@ -71,4 +71,3 @@ rl_agent:
     m_batch_size: 20
     n_epochs: 3
     clip_range: 0.22
-

training/scripts/train_agent.py

@@ -9,9 +9,14 @@
 from tools.model_utils import init_callbacks, get_ppo_instance
 from tools.env_utils import init_envs
 
+def on_shutdown(model):
+    model.env.close()
+    sys.exit()
+
 
 def main():
     args, _ = parse_training_args()
+
     config = load_config(args.config)
 
     populate_ros_configs(config)
@@ -47,7 +52,11 @@ def main():
     eval_cb = init_callbacks(config, train_env, eval_env, PATHS)
     model = get_ppo_instance(config, train_env, PATHS, AgentFactory)
 
-    rospy.on_shutdown(model.env.close())
+    rospy.on_shutdown(lambda: on_shutdown(model))
+
+    ## Save model once
+    if not config["debug_mode"]: 
+        model.save(os.path.join(PATHS["model"], "best_model"))
 
     # start training
     start = time.time()

training/tools/general.py

@@ -220,7 +220,7 @@ def generate_agent_name(config: dict) -> str:
     :param config (dict): Dict containing the program arguments
     """
     if config["rl_agent"]["resume"] is None:
-        START_TIME = dt.now().strftime("%Y_%m_%d__%H_%M")
+        START_TIME = dt.now().strftime("%Y_%m_%d__%H_%M_%S")
         robot_model = rospy.get_param("robot_model")
         architecture_name, encoder_name = config["rl_agent"][
             "architecture_name"

training/tools/model_utils.py

@@ -1,4 +1,5 @@
 import os
+import sys
 from typing import Union, Type
 
 import wandb
@@ -144,9 +145,7 @@ def instantiate_new_model(
         "n_epochs": ppo_config["n_epochs"],
         "clip_range": ppo_config["clip_range"],
         "tensorboard_log": PATHS["tb"],
-        "use_wandb": False
-        if config["debug_mode"]
-        else config["monitoring"]["use_wandb"],
+        # "use_wandb": False if config["debug_mode"] else config["monitoring"]["use_wandb"],
         "verbose": 1,
     }