Quantum Vision for Industrial Autonomy
Computer vision is the perception layer of construction autonomy. Yet mainstream approaches—high-capacity CNNs and Vision Transformers trained on large-scale datasets—often exhibit brittle generalization under domain shift (lighting, texture, camera pose, surface moisture, or sensor differences). They are data-hungry, parameter-intensive, and compute-bound, which limits deployment on edge devices typical of field robotics and mobile inspection platforms.
In our new work, we introduce Hybrid Quantum-Classical Convolutional Models (HQCCMs) — a new class of models that combine quantum feature extraction with deep learning to process images more efficiently. We tested them and found that HQCCMs not only boost accuracy by ~3% over classical CNNs, but also do so with fewer trainable parameters，while reducing parameterization, indicating improved sample efficiency and generalization to previously unseen imagery.
Takeaway: Quantum-enhanced perception is emerging as a practical lever for scaling industrial autonomy—delivering stronger generalization with tighter compute and memory budgets.
Paper: Niu, Y., & Wang*, C. (2025). Hybrid quantum-classical convolutional models for image‐based infrastructure inspection and assessment. Computer-Aided Civil and Infrastructure Engineering. https://doi.org/10.1111/mice.70038
···
