Summary

EasyCarla-RL provides a streamlined, beginner-friendly OpenAI Gym-compatible environment for reinforcement learning (RL) within the CARLA autonomous driving simulator. It significantly reduces the engineering overhead typically associated with integrating complex simulators, enabling researchers and newcomers to efficiently train and evaluate RL agents for autonomous driving tasks.

How It Works

The core of EasyCarla-RL is its lightweight wrapper around the CARLA simulator, exposing a standard Gym API. It integrates essential observation components, including LiDAR scans, ego vehicle states, nearby vehicle data, and waypoints, into a unified, flattened vector. The environment supports safety-aware learning through explicit reward and cost signals and allows customization of traffic density and sensor parameters, facilitating robust agent development.

Quick Start & Requirements

Installation involves cloning the repository, installing dependencies via pip install -r requirements.txt, and then installing the package locally with pip install -e .. A prerequisite is a running CARLA simulator server, with detailed setup instructions available in the official CARLA documentation. A basic interaction demo can be run with python easycarla_demo.py, and advanced usage with pre-trained agents is demonstrated via example/run_dql_in_carla.py.

Highlighted Details

A notable feature is the release of a ~2.76 GB offline dataset in HDF5 format, containing over 7,000 trajectories and 1.1 million timesteps from a mix of expert and random policies. Observations are standardized into a 307-dimensional vector by concatenating ego state, lane info, LiDAR, nearby vehicles, and waypoints. The environment explicitly supports safety-aware learning with cost signals and provides an example for evaluating pre-trained Diffusion Q-Learning agents.

Maintenance & Community

The project is authored by SilverWings and acknowledges contributions from CARLA and gym-carla. Specific details on ongoing maintenance, community channels (like Discord/Slack), or a public roadmap are not detailed in the provided README.

Licensing & Compatibility

EasyCarla-RL is distributed under the permissive Apache License 2.0. This license generally allows for commercial use and integration into closed-source projects without significant restrictions.

Limitations & Caveats

The primary dependency on a separate, potentially complex CARLA simulator installation remains a barrier for users unfamiliar with the CARLA ecosystem. While the RL interface is simplified, the underlying simulator setup requires careful configuration. The large dataset size necessitates substantial download and storage resources.