EdgeNav-QE: QLoRA Quantization and Dynamic Early Exit for LAM-based Navigation on Edge Devices
arXiv:2602.15836v1 Announce Type: cross Abstract: Large Action Models (LAMs) have shown immense potential in autonomous navigation by bridging high-level reasoning with low-level control. However, deploying these multi-billion parameter models on edge devices remains a significant challenge due to memory constraints and latency requirements. In this paper, we propose EdgeNav-QE, a novel framework that integrates Quantized Low-Rank Adaptation (QLoRA) with a dynamic early-exit (DEE) mechanism to optimize LAMs for real-time edge navigation. By quantizing the backbone to 4-bit precision and strategically placing early-exit branches, we enable the model to terminate inference early for simple navigation tasks while retaining full depth for complex decision-making. Experimental results on the Habitat-Sim environment with Matterport3D dataset using OpenVLA-7B backbone, demonstrate that EdgeNav-QE reduces inference latency by 82.7% and memory footprint by 66.7% compared to full-precision base
arXiv:2602.15836v1 Announce Type: cross Abstract: Large Action Models (LAMs) have shown immense potential in autonomous navigation by bridging high-level reasoning with low-level control. However, deploying these multi-billion parameter models on edge devices remains a significant challenge due to memory constraints and latency requirements. In this paper, we propose EdgeNav-QE, a novel framework that integrates Quantized Low-Rank Adaptation (QLoRA) with a dynamic early-exit (DEE) mechanism to optimize LAMs for real-time edge navigation. By quantizing the backbone to 4-bit precision and strategically placing early-exit branches, we enable the model to terminate inference early for simple navigation tasks while retaining full depth for complex decision-making. Experimental results on the Habitat-Sim environment with Matterport3D dataset using OpenVLA-7B backbone, demonstrate that EdgeNav-QE reduces inference latency by 82.7% and memory footprint by 66.7% compared to full-precision baselines, while maintaining 81.8% navigation success rate. Furthermore, it outperforms state-of-the-art static early-exit method by 17.9% in latency, demonstrating the superiority of content-aware adaptive computation for safety-critical applications.
Executive Summary
The article proposes EdgeNav-QE, a framework that integrates QLoRA and dynamic early exit for optimizing Large Action Models (LAMs) on edge devices. It achieves significant reductions in inference latency and memory footprint while maintaining a high navigation success rate. The framework demonstrates the potential for real-time edge navigation in safety-critical applications, outperforming state-of-the-art static early-exit methods.
Key Points
- ▸ Integration of QLoRA and dynamic early exit for LAM optimization
- ▸ Significant reductions in inference latency and memory footprint
- ▸ High navigation success rate maintained
- ▸ Outperformance of state-of-the-art static early-exit methods
Merits
Efficient Model Optimization
The proposed EdgeNav-QE framework efficiently optimizes LAMs for edge devices, enabling real-time navigation
Low Latency and Memory Footprint
The framework achieves significant reductions in inference latency and memory footprint, making it suitable for safety-critical applications
Demerits
Limited Generalizability
The framework's performance may be limited to specific environments and datasets, requiring further testing for broader applicability
Expert Commentary
The EdgeNav-QE framework represents a significant advancement in optimizing LAMs for edge devices. By integrating QLoRA and dynamic early exit, the framework achieves a balance between model efficiency and accuracy. However, further research is necessary to ensure the framework's generalizability and robustness in various environments and applications. The implications of this work are far-reaching, with potential applications in autonomous navigation, edge AI, and beyond.
Recommendations
- ✓ Further testing and evaluation of the EdgeNav-QE framework in diverse environments and datasets
- ✓ Exploration of the framework's potential applications in various industries, including autonomous vehicles and robotics
