[Paper Review] A Software Architecture for Autonomous Vehicles: Team LRM-B Entry in the First CARLA Autonomous Driving Challenge

Luis Alberto Rosero, Iago Pachêco Gomes+6 more|arXiv (Cornell University)|Oct 23, 2020

Autonomous Vehicle Technology and Safety84 references11 citations

TL;DR

This paper presents a modular, robust software architecture for autonomous vehicles developed by Team LRM-B for the CARLA Autonomous Driving Challenge, integrating 3D obstacle detection, CNN-based traffic sign recognition, risk assessment via short-term motion prediction, Markov Decision Process (MDP) for decision-making, and Model Predictive Control (MPC) for control. The system achieved first place in three tracks and second in one, demonstrating high reliability in complex urban scenarios with minimal traffic violations.

ABSTRACT

The objective of the first CARLA autonomous driving challenge was to deploy autonomous driving systems to lead with complex traffic scenarios where all participants faced the same challenging traffic situations. According to the organizers, this competition emerges as a way to democratize and to accelerate the research and development of autonomous vehicles around the world using the CARLA simulator contributing to the development of the autonomous vehicle area. Therefore, this paper presents the architecture design for the navigation of an autonomous vehicle in a simulated urban environment that attempts to commit the least number of traffic infractions, which used as the baseline the original architecture of the platform for autonomous navigation CaRINA 2. Our agent traveled in simulated scenarios for several hours, demonstrating his capabilities, winning three out of the four tracks of the challenge, and being ranked second in the remaining track. Our architecture was made towards meeting the requirements of CARLA Autonomous Driving Challenge and has components for obstacle detection using 3D point clouds, traffic signs detection and classification which employs Convolutional Neural Networks (CNN) and depth information, risk assessment with collision detection using short-term motion prediction, decision-making with Markov Decision Process (MDP), and control using Model Predictive Control (MPC).

Motivation & Objective

• To design a scalable and robust software architecture for autonomous navigation in complex urban traffic scenarios using simulation.
• To minimize traffic infractions by ensuring compliance with traffic laws in diverse driving situations.
• To enable autonomous operation across multiple tracks with varying sensor configurations and environmental conditions.
• To extend the CaRINA 2 platform with enhanced perception, risk assessment, and decision-making components for real-world simulation.
• To validate the architecture in a large-scale, multi-scenario simulation environment using the CARLA simulator.

Proposed method

• Employed a modular architecture based on CaRINA 2, integrating perception, planning, and control modules.
• Used 3D point clouds from LiDAR and stereo vision for obstacle detection via height maps and clustering.
• Applied Convolutional Neural Networks (CNNs) with depth information for accurate traffic sign detection and classification.
• Implemented short-term motion prediction for real-time risk assessment and collision detection using Distance-to-Collision (DTC).
• Utilized Markov Decision Processes (MDP) for high-level decision-making under uncertainty in dynamic traffic environments.
• Applied Model Predictive Control (MPC) for smooth and safe trajectory tracking and control.

Experimental results

Research questions

• RQ1 How can a modular software architecture be designed to support autonomous navigation across diverse, complex urban traffic scenarios in simulation?
• RQ2 What perception and decision-making components are most effective for minimizing traffic violations in simulated urban driving?
• RQ3 How does integrating short-term motion prediction improve risk assessment and collision avoidance?
• RQ4 To what extent can MDP-based decision-making handle varied traffic conditions such as traffic jams, signalized intersections, and merging maneuvers?
• RQ5 How does the fusion of 2D detection with 3D point cloud projection enhance object localization and decision accuracy?

Key findings

• The LRM-B team achieved first place in three out of four tracks of the CARLA Autonomous Driving Challenge, with a second-place finish in the remaining track.
• The system successfully navigated complex urban scenarios for several hours, demonstrating robustness and reliability in real-time simulation.
• The integration of 2D object detection with 3D point cloud projection significantly improved the accuracy of obstacle, traffic light, and sign localization.
• The use of short-term motion prediction and Distance-to-Collision (DTC) enhanced real-time risk assessment, reducing collision incidents.
• The MDP-based decision-making framework effectively managed diverse traffic scenarios, including intersections, merging, and traffic jams.
• The modular architecture enabled rapid extension and adaptation to new tracks, supporting diverse sensor configurations and requirements.