Precision Switching Schedule for Efficient Control Implementations

TL;DR

This paper introduces a novel precision switching schedule for feedback controllers in cyber-physical systems, optimizing the trade-off between control quality and computational efficiency by switching between floating-point precisions.

Contribution

It formulates the precision switching problem as a multi-objective MIQP and provides an efficient method to compute schedules that balance control performance and runtime.

Findings

Switching between 32-bit and 16-bit floating-point reduces runtime by 26.5%.

Control performance improves by 27.6% over 16-bit execution.

The approach maintains system output within a specified reference band.

Abstract

Modern cyber-physical systems, such as automotive control, rely on feedback controllers that regulate the system towards desired a setpoint. In practice, however, the controller must also be scheduled efficiently on resource-constrained processors, where the choice of numerical precision for controller implementation directly affects both control quality and computational cost. This trade-off is critical: higher precision improves control performance but increases runtime, while lower precision executes faster in the processor but may degrade overall system performance. In this work, we propose the first approach for a precision switching schedule, where the controller switches between different floating-point precisions to balance control performance and enhance computational efficiency. We formulate this problem as a multi-objective optimization, expressed as a Mixed-Integer Quadratic Program (MIQP) with sound linearizations and error bounds that capture roundoff effects from different precision implementations. Our method efficiently computes a switching schedule that ensures the system output remains within a specified reference band. Through experimental evaluation on standard benchmark control systems, we demonstrate that switching between 32-bit and 16-bit floating-point implementations offers an average runtime reduction of 26.5% compared to 32-bit execution and a 27.6% improvement in control performance over 16-bit execution, while maintaining near-optimal overall performance.