This repository provides an OpenAI Gym wrapper for the CARLA simulator, enabling reinforcement learning agents to interact with a realistic driving environment. It is targeted at researchers and developers in autonomous driving and deep reinforcement learning, offering a standardized interface for training and evaluating RL algorithms in a simulated urban setting.

How It Works

The wrapper integrates with CARLA, a popular open-source simulator for autonomous driving research. It exposes CARLA's sensor data (camera, lidar, semantic segmentation) and vehicle state as observations within the Gym API. The environment defines termination conditions like collisions or reaching a destination and offers a customizable reward function based on speed, lane adherence, and penalties.

Quick Start & Requirements

• Install via pip: pip install -r requirements.txt and pip install -e .
• Requires Python 3.6 and CARLA 0.9.6.
• CARLA must be added to PYTHONPATH.
• CARLA server launch: ./CarlaUE4.sh -windowed -carla-port=2000
• Official documentation: Not explicitly linked, but usage is detailed in test.py.

Highlighted Details

• Provides dictionary observations including camera, lidar, and semantic representations.
• Offers a state vector observation with lateral distance, heading error, speed, and front vehicle proximity.
• Customizable termination conditions and reward functions.
• Links to related repositories for deep reinforcement learning and imitation learning in autonomous driving.

Maintenance & Community

• Maintained by cjy1992.
• No explicit community links (Discord, Slack) or roadmap are provided in the README.

Licensing & Compatibility

• The README does not specify a license.

Limitations & Caveats

The environment is specifically built for CARLA version 0.9.6 and Ubuntu 16.04, potentially limiting compatibility with newer CARLA versions or other operating systems. The lack of explicit licensing information may pose a concern for commercial use.