WorldOnRails

Overview:

The "World on Rails" repository contains the code for the technical report "Learning to drive from a world on rails" by Dian Chen, Vladlen Koltun, and Philipp Krähenbühl. The repository provides resources for training agents to drive in the CARLA simulator environment. The code includes pre-computed Q values in the dataset along with instructions for setting up the environment and training the models. The content focuses on training agents for driving tasks using machine learning techniques.

Features:

• Pre-computed Q values: Includes pre-computed Q values in the dataset for driving tasks.
• CARLA Setup: Provides instructions for setting up CARLA and training the models.
• Training Guides: Includes guides for training the World-on-Rails and LBC agents.
• Evaluation Instructions: Guidelines for evaluating the pretrained weights with specific launch configurations.
• Leaderboard Routes: Provides routes for evaluating performance with options for different agents.
• Dataset Release: Offers the data trained for the leaderboard tasks.
• Acknowledgements: Credits original sources for leaderboard and scenario runner codes.
• License: Released under the MIT License along with references to other licenses.

Summary:

The "World on Rails" repository offers a comprehensive resource for training driving agents in a simulated environment. It provides pre-computed data, training guides, and evaluation instructions, catering to researchers and developers interested in machine learning techniques applied to autonomous driving tasks. Proper setup and adherence to the installation guidelines will enable users to effectively train and evaluate agents within the CARLA simulator. The inclusion of acknowledgements and licensing information ensures proper credit and compliance with intellectual property rights.