Accurate object localization is essential for enabling autonomous operation of indoor robotic systems. As a low-cost, compact, and flexibly deployable solution, monocular visual object localization (MVOL) is highly applicable to lightweight embedded robotic platforms. However, conventional data-driven MVOL methods suffer from inherent limitations of 2D-to-3D ill-posed mapping, which is caused by the incapability of constraining the spatial physical logic of real scenes, resulting in severe depth ambiguity, inaccurate scale estimation, and physically unreasonable predictions. To address these issues, this paper proposes a novel physics-guided monocular visual localization framework termed PC-IMVL for indoor scenarios. The PC-IMVL integrates deep visual perception with embedded physical modeling, which explicitly introduces spatial physical constraints into the network optimization process and builds a physical consistency-aware loss function to regularize 3D position and pose estimation. Combined with a lightweight tailored architecture, the framework enables efficient and reliable embedded deployment. Offline experiments and real-world online tests validate the effectiveness of the proposed method. PC-IMVL yields average absolute errors (AE) of 0.095–0.333 m, reducing the localization error of early fusion methods by more than 50%. Within a working distance of 3–4 m, it achieves a relative error (RE) of 2.4% and a horizontal viewing angle error (VAE) below 2°, outperforming existing state-of-the-art MVOL methods. The effectiveness of the physical guidance mechanism is verified. This work provides a practical high-precision localization solution for embedded indoor robotic systems.
Haorui Ge, Luzheng Bi, Weijie Fei et al.· Italian National Conference...· 0 citations
Mechanistic simulation and machine learning are powerful but complementary tools: physics-based simulation is interpretable yet computationally expensive and blind to real-world context, whereas machine learning is fast but data-hungry and opaque. Biological systems resolve this tension elegantly, coupling physically grounded mechanoreceptor sensing with higher-level neural interpretation that places those signals in context. Inspired by this layered architecture, we present Mechanics-AI, an open-source framework that mirrors the same sensing-then-interpretation logic computationally. A first learning layer (ML1) emulates expensive finite-element, computational fluid dynamics and multiphysics simulations to produce interpretable physical metrics such as stress, strain, shear, and thermal and moisture fields, while a second layer (ML2) fuses these metrics with heterogeneous real-world metadata to predict categorical outcomes and design recommendations. Eight algorithms are benchmarked automatically, the most accurate is selected for each task, and Shapley additive explanations expose the dominant physical drivers to preserve interpretability. The framework is demonstrated across three independent domains using a single unchanged pipeline: forensic traumatic brain injury prediction, optimisation of a bio-inspired humanoid bioreactor for tissue engineering, and a zero-emission building (ZEBAI) framework that couples thermo-hygro-mechanical simulation with Sobol-sampled surrogate modelling to design sustainable, low-carbon envelopes from recycled aggregate concrete by balancing structural safety, energy and embodied carbon. Despite entirely different physics, data and objectives, the same architecture generalises across all three, showing that bio-inspired, layered coupling of mechanistic simulation and contextual learning offers a reusable, interpretable route to cross-domain engineering prediction and sustainable design.
Yuyang Wei, Weijie Fei, Jiarong Wang et al.· Biomimetics· 0 citations