Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Analysis of a Modular Autonomous Driving Architecture: The Top Submission to CARLA Leaderboard 2.0 Challenge

Weize Zhang, Mohammed Elmahgiubi, Kasra Rezaee, Behzad Khamidehi, Hamidreza Mirkhani, Fazel Arasteh, Chunlin Li, Muhammad Ahsan Kaleem, Eduardo R. Corral-Soto, Dhruv Sharma, Tongtong Cao

TL;DR

This work presents Kyber-E2E's modular autonomous driving architecture for CARLA Leaderboard 2.0, combining sensing, localization, perception, tracking/prediction, and planning/control to win the map track. It leverages language-assisted perception and IRL-based tuning of the motion planner, trained on diverse open datasets (including inD) and supplemented by CARLA Town data, to achieve robust performance across complex traffic scenarios. The study provides a detailed ablation of module contributions, showing where resource allocation yields the most benefit and highlighting dependence on accurate perception for planning in high-stakes maneuvers. The results demonstrate the practicality of modular designs for rapid development and deployment, while noting current limitations in long-range perception and the potential gains from moving toward end-to-end approaches in future work.

Abstract

In this paper we present the architecture of the Kyber-E2E submission to the map track of CARLA Leaderboard 2.0 Autonomous Driving (AD) challenge 2023, which achieved first place. We employed a modular architecture for our solution consists of five main components: sensing, localization, perception, tracking/prediction, and planning/control. Our solution leverages state-of-the-art language-assisted perception models to help our planner perform more reliably in highly challenging traffic scenarios. We use open-source driving datasets in conjunction with Inverse Reinforcement Learning (IRL) to enhance the performance of our motion planner. We provide insight into our design choices and trade-offs made to achieve this solution. We also explore the impact of each component in the overall performance of our solution, with the intent of providing a guideline where allocation of resources can have the greatest impact.

Analysis of a Modular Autonomous Driving Architecture: The Top Submission to CARLA Leaderboard 2.0 Challenge

TL;DR

This work presents Kyber-E2E's modular autonomous driving architecture for CARLA Leaderboard 2.0, combining sensing, localization, perception, tracking/prediction, and planning/control to win the map track. It leverages language-assisted perception and IRL-based tuning of the motion planner, trained on diverse open datasets (including inD) and supplemented by CARLA Town data, to achieve robust performance across complex traffic scenarios. The study provides a detailed ablation of module contributions, showing where resource allocation yields the most benefit and highlighting dependence on accurate perception for planning in high-stakes maneuvers. The results demonstrate the practicality of modular designs for rapid development and deployment, while noting current limitations in long-range perception and the potential gains from moving toward end-to-end approaches in future work.

Abstract

In this paper we present the architecture of the Kyber-E2E submission to the map track of CARLA Leaderboard 2.0 Autonomous Driving (AD) challenge 2023, which achieved first place. We employed a modular architecture for our solution consists of five main components: sensing, localization, perception, tracking/prediction, and planning/control. Our solution leverages state-of-the-art language-assisted perception models to help our planner perform more reliably in highly challenging traffic scenarios. We use open-source driving datasets in conjunction with Inverse Reinforcement Learning (IRL) to enhance the performance of our motion planner. We provide insight into our design choices and trade-offs made to achieve this solution. We also explore the impact of each component in the overall performance of our solution, with the intent of providing a guideline where allocation of resources can have the greatest impact.
Paper Structure (26 sections, 3 figures, 2 tables)

This paper contains 26 sections, 3 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Agent Architecture
  3. Sensing
  4. Perception
  5. Signs Detection and Recognition
  6. Object Detection
  7. Post-processing
  8. Tracking and prediction
  9. Tracking
  10. Prediction
  11. Planning and Control
  12. Behavior planning
  13. Motion Planning
  14. Controller
  15. Experiments and result
  16. ...and 11 more sections

Figures (3)

  • Figure 1: Kyber-E2E agent architecture.
  • Figure 2: perception modules.
  • Figure 3: Tracking/prediction module