Multi-Objective Optimization of ML Algorithms and FPGA Accelerators for Robotic Intelligence

Abstract

Integrating machine learning (ML) algorithms into robotic intelligence necessitates balancing accuracy, resource efficiency, and energy consumption. This study presents the Multi-Objective Optimization Co-Design (MOO-CoD) framework, developed to enhance both ML algorithms and their corresponding hardware accelerators for robotic applications. MOO-CoD simultaneously optimizes algorithms and hardware through MOO techniques, including weighted optimization, accuracy-prioritized approaches, and Pareto front analysis, to achieve optimal model performance and resource efficiency. By leveraging FPGA accelerators, the MOO-CoD framework enables rapid ML deployment with scalable, high-performance hardware designs, making it ideal for low-latency and energy-efficient robotic systems. Empirical evaluations using the NGSIM and SECOM datasets demonstrate that MOO-CoD significantly reduces resource consumption while maintaining high AUC scores. These results highlight the framework’s effectiveness in addressing critical challenges in robotic intelligence by enhancing both ML performance and hardware efficiency.