Recti-Q: Feature-Space Rectification for Out-of-Distribution-Robust Quantized Perception in Edge Robotics
Quick Answer
Recti-Q introduces a lightweight feature-space rectification framework to enhance out-of-distribution robustness in quantized perception for edge robotics.
Quick Take
It effectively recovers lost robustness in 4-bit PTQ models, achieving performance comparable to FP32 with minimal parameter overhead and negligible compute costs, making it suitable for deployment in unpredictable environments.
Key Points
- Recti-Q improves robustness in 4-bit PTQ models without significant in-distribution accuracy loss.
- The framework is architecture-agnostic, applicable to both CNNs and Transformers.
- It adds less than 1% parameter overhead while preserving over 99% of PTQ memory savings.
- Enables low-bandwidth OTA resilience patching for robotic fleets in unpredictable environments.
- Accepted for presentation at IROS 2026, highlighting its relevance in robotics.
DeepSignal Analysis
What happened
Recti-Q is a proposed framework aimed at improving the robustness of quantized perception models in edge robotics. It specifically addresses the challenges posed by out-of-distribution scenarios, where traditional post-training quantization (PTQ) methods fall short. The framework is designed to work with existing architectures while maintaining low computational overhead.
Key evidence
- 4-bit PTQ models show significant robustness degradation under distribution shifts, despite maintaining in-distribution accuracy, as evidenced by benchmarks like ImageNet-C and PACS.
- Recti-Q employs a lightweight feature-space rectification approach that freezes the quantized backbone and trains a small classifier-head LoRA adapter using only source data.
- The framework achieves less than 1% parameter overhead while preserving over 99% of PTQ memory savings, making it suitable for deployment in unpredictable environments.
Why it matters
The introduction of Recti-Q is significant for edge robotics, where reliability in diverse conditions is crucial. Traditional PTQ methods often fail to ensure robustness under real-world variations, which can lead to performance issues. By addressing this gap, Recti-Q enhances the applicability of quantized models in practical scenarios, potentially improving the deployment of robotic systems in challenging environments.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Robotic perception pipelines increasingly rely on large vision backbones deployed on SWaP-constrained edge platforms, making post-training quantization (PTQ) attractive for real-time inference. However, while PTQ often preserves clean in-distribution accuracy, we show that it can substantially degrade reliability under deployment-relevant distribution shifts (e.g., sensor noise, severe weather, and novel operating environments), creating a Quantization-Induced Robustness Gap. Across foundational vision benchmarks (ImageNet-C and PACS), 4-bit PTQ models exhibit pronounced robustness degradation despite negligible ID accuracy loss. To address this, we propose Recti-Q, a lightweight feature-space rectification framework that freezes the quantized backbone and trains a small classifier-head LoRA adapter using only source data. Recti-Q is architecture-agnostic across CNNs and Transformers, supports efficient teacher-free training, and recovers a significant portion of the lost robustness, in some cases matching or exceeding FP32 performance. At less than 1% parameter overhead (as small as 6 KB), Recti-Q preserves over 99% of PTQ memory savings, adds negligible compute, and enables low-bandwidth Over-The-Air (OTA) resilience patching for deployed robotic fleets operating in unpredictable physical environments.
| Comments: | Accepted at the 2026 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2026) |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG); Robotics (cs.RO) |
| Cite as: | arXiv:2607.18540 [cs.CV] |
| (or arXiv:2607.18540v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2607.18540 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Hamidreza Yaghoubi Araghi [view email]
[v1]
Mon, 20 Jul 2026 22:14:04 UTC (951 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.