Deep Reinforcement Learning (DRL) has achieved remarkable success in robotic control, autonomous driving, and game-playing agents. However, its decision-making process often remains a black box, lacking both interpretability and verifiability. In robotic control tasks, developers cannot pinpoint decision errors or precisely adjust control strategies based solely on observed robot behaviors. To address this challenge, this work proposes an interpretable DRL framework based on a Causal Correction and Compensation Network (C2-Net), which systematically captures the causal relationships underlying decision-making and enhances policy robustness. C2-Net integrates a Graph Neural Network-based Neural Causal Model (GNN-NCM) to compute causal influence weights for each action. These weights are then dynamically applied to correct and compensate the raw policy outputs, thereby balancing performance optimization and transparency. This work validates the approach on OpenAI Gym’s Hopper, Walker2d, and Humanoid environments, as well as the multi-agent AzureLoong platform built on Isaac Gym. In terms of convergence speed, final return, and policy robustness, experimental results show that C2-Net achieves higher performance over both non-causal baselines and conventional attention-based models. Moreover, it provides rich causal explanations for its decisions. The framework represents a principled shift from correlation to causation and offers a practical solution for the safe and reliable deployment of multi-robot systems.
During the operation of steel cable-driven radiation-resistant robots in nuclear industrial environments, the tensile force of a steel cable is influenced by temperature variations, which can cause significant detection errors. To address this problem, this study proposes a temperature-compensated axial force characterization method for steel cables based on the magnetoelastic effect, aiming to ensure the measurement accuracy of magnetoelastic sensors. The principle of the magnetoelastic measurement method involves magnetizing the steel cable. When subjected to tensile forces, the magnetization characteristics of the steel cable change, thereby altering the detection signal of the magnetoelastic sensor. By analyzing the relationship between steel cable tension and variations in the detection signal, effective force measurement can be achieved. First, experiments are conducted to investigate the influence of temperature on the detection signals of a magnetoelastic sensor under zero-load conditions. Then, additional tests are performed to examine the combined effects of a tensile force and temperature on the sensor’s signals. Finally, based on the experimental data, axial force prediction models are constructed using both surface fitting and a backpropagation neural network (BPNN). The results demonstrate that, compared to the resistance values, inductance exhibits superior stability under temperature variations. In the temperature range of 20–50 °C, the inductance variation is approximately 0.15 μH, which indicates improved suitability for characterizing the axial force of steel cables. It is also shown that under isothermal conditions, the inductance increases linearly with the applied tensile force, exhibiting a slope of approximately 0.025 μH/kN. Both the surface fitting-based and BPNN-based axial force prediction models demonstrate high accuracy, with absolute prediction errors consistently below 5% compared to actual data.
With the increasing application of oxide films in nuclear fuel assemblies, the accurate measurement of thin-film thickness has become increasingly critical. Traditional spectral interferometry techniques have limitations when dealing with new materials and complex structures; therefore, this study proposes a multi-channel wide-spectrum high-resolution analysis technique. This technique optimizes the utilization of photosensitive elements through multi-channel spectral sampling, combined with precision spectroscopic components and an independent optical focusing and imaging system. Simultaneously, it adopts optical correction technologies such as coma optimization and astigmatism correction to improve imaging quality and spectral resolution. Additionally, it enhances data accuracy by means of multi-channel calibration based on the least squares method and non-linear correction. The technique enables high-precision measurement ranging from the nanometer to the millimeter scale, resulting in a significantly wider measurement range compared to traditional spectrometers. Simulation verification shows that this technique outperforms existing technologies in information acquisition, analysis accuracy, and detection efficiency, and has broad application prospects in fields such as semiconductor chip manufacturing and optical coating. In the future, focus will be placed on expanding the spectral range, improving resolution, and enhancing real-time measurement capabilities.
In the context of the increasing demands of precision manufacturing and nanotechnology, especially for emerging fields such as Oxide oxide films in Nuclear nuclear fuel assemblies, the measurement of multi-layer inhomogeneous thin films faces significant challenges. Traditional spectroscopic interference thickness measurement techniques have limitations in handling dispersion interference, parameter coupling, and the efficient solution of nonlinear inverse problems. This study proposes a new model that integrates deep learning and physical model fitting. It constructs a theoretical model of multi-layer thin-film interference spectroscopy based on the Lorentz–Drude formula, uses a generative adversarial network (GAN) for initial structure analysis, and builds a two-layer optimization framework of “deep learning rough positioning—physical model fine fitting”. The research aims to break through the limitations of traditional methods, improve measurement accuracy and anti-noise ability, and provide a key technical support for emerging fields.