Optimization of a Line Detection Algorithm for Autonomous Vehicles on a RISC-V with Accelerator

Abstract

This paper explores the optimization of a line detection algorithm critical for autonomous vehicles, addressing strict real-time and safety requirements often unmet by general-purpose processors alone. The approach utilizes a heterogeneous system featuring a general-purpose RISC-V core coupled with a domain-specific hardware accelerator. By analyzing and adapting computationally intensive code sections for the accelerator, the system successfully verified real-time performance, yielding a notable 3.7x speedup.

Report

Key Highlights

• Target Application: Line detection algorithm optimization specifically designed for autonomous vehicle environments.
• Hardware Strategy: Implementation on a heterogeneous system comprising a general-purpose RISC-V core and a domain-specific accelerator.
• Performance Gain: The accelerated system achieved a significant 3.7x speedup compared to running the application solely on the RISC-V core (unaccelerated).
• Motivation: Meeting the stringent safety, reliability, and real-time processing requirements essential for autonomous driving decision making.

Technical Details

• Architecture: The core platform is a heterogeneous computing system, utilizing the flexibility of a RISC-V processor alongside specialized hardware.
• Methodology: The optimization focused on analyzing the application code to identify and adapt the most computationally intensive sections, which were subsequently offloaded to the domain-specific accelerator.
• Requirements: The solution was verified against the strict real-time operational requirements characteristic of autonomous vehicle safety systems.

Implications

• RISC-V Adoption in Critical Systems: This work validates RISC-V as a suitable and capable architecture for highly demanding, safety-critical edge applications, such as Advanced Driver Assistance Systems (ADAS) and autonomous vehicles.
• Accelerator Integration: It reinforces the necessity and effectiveness of pairing RISC-V cores with custom, domain-specific accelerators to achieve necessary real-time throughput and latency targets in complex machine vision tasks.
• Ecosystem Development: The successful optimization and demonstrable speedup contribute practical evidence and blueprints for implementing high-performance, custom hardware solutions within the growing RISC-V ecosystem for specialized domains.