--- layout: home title: CARLA Leaderboard description: CARLA Autonomous Driving competition. background: '/img/carla_header.png' image: 'img/carla_header.png' ---

Introduction

The CARLA leaderboard is a way to...

Participation tracks

The challenge will have four parallel tracks, which differ in the type of input data provided to the agents. In all tracks, the agents will have access to additional measurements, such as a high-level plan of how to reach destinations, and the current speed.

Competition metrics

The performance of an agent will depend on the number of routes successfully completed. A route is considered successfully completed if no critical infractions were triggered. If the agent triggers a critical infraction, the episode will be automatically terminated and the agent will only get a score proportional to the percentage of the route it completed until the critical infraction. The list of critical infractions is as follows: Hitting the static scenery (poles, traffic signs, etc.). If an agent runs out of time to complete a given route the final score will be the percentage of route completed before the timeout. The final score will consist of two components: Points obtained for route completion. Points lost due to other infractions (e.g., running a red light). Minor infractions will have a negative effect on the final score. Each infraction discounts points from the final score, which could get down to 0 in the worst case. These are the points discounted for each infraction:

Participation mechanics

Participants must create a new team in the EvalAI CARLA AD Challenge contest. Teams will create a docker image using the provided Dockerfile templates. Docker containers will be then pushed to CARLA AD Challenge internal repositories using EvalAI cli scripts. Teams must select the right challenge phase to submit their containers. Docker images will be automatically evaluated in AWS by our evaluation infrastructure and results will be reported to users through the EvalAI CARLA AD Challenge site. For detailed instructions, please visit the Submit section. Important considerations Agents will run on a private AWS instance endowed with a multi-processor and a GPU (NVIDIA Tesla K80). Teams will receive a budget of 300 hours for the entire challenge. If a team exceeds its hours, new docker containers won’t be evaluated. The CARLA AD Challenge organization reserves the right to assign further hours to teams due to extraordinary circumstances. Some Autonomous Driving baselines are provided as a reference. Please, read the Get started section for more information. We recommend participants to join our Discord chat to keep up-to-date with challenge news.