This is the error message. I'm not sure, but it looks like I can't download rollout_rl_agent.launch from the minio bucket maybe? The other file (training_params.yaml) is there and should be downloading fine.

The ROS UI comes up in VNC, but it doesn't have a model loaded and is running, but nothing really loaded into it.

================================================================================REQUIRED process [download_params_and_roslaunch_agent_node-2] has died!
process has died [pid 110, exit code 1, cmd /opt/install/deepracer_simulation_environment/lib/deepracer_simulation_environment/download_params_and_roslaunch_agent.py us-east-1 bucket current custom_files/training_params.yaml rollout_rl_agent.launch __name:=download_params_and_roslaunch_agent_node __log:=/root/.ros/log/f06930f0-9232-11eb-b7b7-02420a0a061a/download_params_and_roslaunch_agent_node-2.log].
log file: /root/.ros/log/f06930f0-9232-11eb-b7b7-02420a0a061a/download_params_and_roslaunch_agent_node-2*.log
Initiating shutdown!

when I try using DEEP_CONVOLUTIONAL_NETWORK (before used DEEP_CONVOLUTIONAL_NETWORK_SHALLOW, It run smoothly).

It raises an error in evaluation step:

Training> Name=main_level/agent, Worker=0, Episode=20, Total reward=2.1, Steps=99, Training iteration=0
## agent: Starting evaluation phase
Testing> Name=main_level/agent, Worker=0, Episode=20, Total reward=1.5, Steps=99, Training iteration=0
## agent: Finished evaluation phase. Success rate = 0.0, Avg Total Reward = 1.5
## agent: Starting evaluation phase
Exception in thread Thread-435:
Traceback (most recent call last):
  File "/usr/local/lib/python3.5/dist-packages/redis/connection.py", line 185, in _read_from_socket
    raise socket.error(SERVER_CLOSED_CONNECTION_ERROR)
OSError: Connection closed by server.

During handling of the above exception, another exception occurred:

Traceback (most recent call last):
  File "/usr/lib/python3.5/threading.py", line 914, in _bootstrap_inner
    self.run()
  File "/usr/local/lib/python3.5/dist-packages/redis/client.py", line 3236, in run
    timeout=sleep_time)
  File "/usr/local/lib/python3.5/dist-packages/redis/client.py", line 3135, in get_message
    response = self.parse_response(block=False, timeout=timeout)
  File "/usr/local/lib/python3.5/dist-packages/redis/client.py", line 3036, in parse_response
    return self._execute(connection, connection.read_response)
  File "/usr/local/lib/python3.5/dist-packages/redis/client.py", line 3013, in _execute
    return command(*args)
  File "/usr/local/lib/python3.5/dist-packages/redis/connection.py", line 637, in read_response
    response = self._parser.read_response()
  File "/usr/local/lib/python3.5/dist-packages/redis/connection.py", line 290, in read_response
    response = self._buffer.readline()
  File "/usr/local/lib/python3.5/dist-packages/redis/connection.py", line 224, in readline
    self._read_from_socket()
  File "/usr/local/lib/python3.5/dist-packages/redis/connection.py", line 199, in _read_from_socket
    (e.args,))
redis.exceptions.ConnectionError: Error while reading from socket: ('Connection closed by server.',)

In late July 2020, the AWS DeepRacer changed the requirements on the format of uploaded models. The model format is described at
https://docs.aws.amazon.com/deepracer/latest/developerguide/deepracer-troubleshooting-service-migration-errors.html

The current best way for local training seems to be:

1. Upload model to s3
2. Import model from s3 in AWS DeepRacer console.

Models uploaded with the script upload-current.sh fail to import with the new DeepRacer model service.

I have a I5 3570, which apparently does not have AVX2 capability.
When I try to run the script, something breaks saying that I don't have AVX2 and then everything simply shuts down. I get a new tab on Chrome (I assume supposed to stream the training) but no connection.

I tried this without TensorFlow installed and
Installing it with a simple "pip install tensorflow".
Also tried with tensorflow-gpu (both with and without the normal tensorflow installed).

As none of the above options worked, I assume the repo already have tensorflow somehow, probably something docker related (of which I understand absolutely nothing).

Would be nice to have this being compatible with older machines like mine, which are actually quite common in less fortunate countries (which are, ironically, also places where paying an EC2 instance are quite not a possibility).

I would like to ask, where can I find the file describing the logic of the RL algorithm, I would like to know if other RL algorithms can be used instead of PPO. Another similar question is whether we can customize the structure of the neural network and what files should we modify to achieve this purpose.