build details

Show: section status errors & todos local changes recent changes last change in-page changes feedback controls

Imitation Learning from Logs

Modified 2019-04-15 by Liam Paull

In this part, you can find all the required steps in order to make a submission based on Imitation Learning with Tensorflow for the lane following task using real log data. It can be used as a starting point for any of the LF, LFV, and LFVI challenges.

That you have made a submission with the tensorflow template.

You win the AI-DO!


Modified 2019-04-23 by jzilly

Clone the baseline Tensorflow imitation learning from logs repository:

$ git clone https://github.com/duckietown/challenge-aido_LF-baseline-IL-logs-tensorflow.git
$ cd challenge-aido_LF-baseline-IL-logs-tensorflow

The code you find is structured into 3 folders.

  1. Extracting data (folder extract_data)

  2. Learning from the data (folder learning)

  3. Submitting learned model (folder imitation_agent)

Step 2 can be run either using Docker or without. All other steps require Docker.

Extract data

Modified 2018-10-28 by liampaull

Go to the extract_data folder:

$ cd extract_data


$ make docker_extract_data

Once extraction is completed, move the extracted data to the learning folder. Type:

$ make docker_copy_for_learning


Modified 2018-10-28 by liampaull

Go to the learning folder

$ cd ../learning

Before being able to conduct the learning experiments in docker, you will need the Nvidia-runtime environment. To get this dependency type:

$ make prepare-docker

If using Docker (to avoid a ROS installation), type:

$ make learn-docker

to train a small convolutional neural network for imitation learning. This assumes that you have extracted data in the previous step.

Once learning is completed, move the learned model to the submission folder. Type:

$ make copy_for_submission

If you already have your Tensorflow learning pipeline on your system setup and do not use Docker. Then type:

$ make install-dependencies

To install some dependencies in a Python 2.7 virtual environment.

To train a small convolutional neural network for imitation learning, type:

$ make learn-regular

Once learning is completed, move the learned model to the imitation agent folder. Type:

$ make regular_copy_for_submission


Modified 2019-04-23 by jzilly

Go to the imitation_agent folder

$ cd ../imitation_agent

If you have completed the previous steps, you will be able to submit this as is, with:

$ dts challenges submit

Or, run local evaluation with:

$ dts challenges evaluate

No questions found. You can ask a question on the website.