Saliency-Aware Quantized Imitation Learning for Efficient Robotic Control

TL;DR

This paper introduces SQIL, a saliency-aware quantization method for imitation learning that maintains high decision accuracy in resource-limited robotic applications, enabling faster and more energy-efficient deployment.

Contribution

The paper proposes a novel saliency-aware quantization technique that improves the efficiency of imitation learning models without significant accuracy loss.

Findings

4-bit quantized VLA model achieves 2.5x speedup

Model maintains performance across diverse benchmarks

Significant energy savings on edge GPU

Abstract

Deep neural network (DNN)-based policy models, such as vision-language-action (VLA) models, excel at automating complex decision-making from multi-modal inputs. However, scaling these models greatly increases computational overhead, complicating deployment in resource-constrained settings like robot manipulation and autonomous driving. To address this, we propose Saliency-Aware Quantized Imitation Learning (SQIL), which combines quantization-aware training with a selective loss-weighting strategy for mission-critical states. By identifying these states via saliency scores and emphasizing them in the training loss, SQIL preserves decision fidelity under low-bit precision. We validate SQIL's generalization capability across extensive simulation benchmarks with environment variations, real-world tasks, and cross-domain tasks (self-driving, physics simulation), consistently recovering full-precision performance. Notably, a 4-bit weight-quantized VLA model for robotic manipulation achieves up to 2.5x speedup and 2.5x energy savings on an edge GPU with minimal accuracy loss. These results underline SQIL's potential for efficiently deploying large IL-based policy models on resource-limited devices.