CXChengcheng Xu

Chengcheng Xu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Chengcheng Xu has a total of 17 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Multi-task engine knock detection and intensity estimation based on temporal convolutional networks

    by Pengyuan Sun, Huichao Zhao, Hanshi Qu, Tianyu Gao, Weixuan Chen, Chengcheng Xu

    To address the challenge of reliable engine knock detection under diverse operating conditions, this paper proposes a multi-task framework based on a multi-head temporal convolutional network (TCN) for simultaneous detection and intensity estimation. In-cylinder pressure signals are transformed into a multi-channel representation through Savitzky–Golay detrending and engine-speed-adaptive angle-domain bandpass filtering to suppress low-frequency combustion trends and enhance oscillatory knock-related components. The proposed framework employs a multi-head TCN to regress knock intensity, quantified by the root mean square of the bandpass-filtered residual within a spark-relative or adaptive sliding window, and to classify knock occurrence, enabling a comprehensive evaluation of combustion stability. An operating-condition-adaptive multi-feature labeling strategy with robust normalization and Otsu-guided thresholding is adopted to ensure consistent supervision across varying engine states. To alleviate the adverse effects of class imbalance between knock and non-knock cycles, weighted sampling and class-balanced loss functions are incorporated to improve training stability and generalization. Experimental evaluations across multiple engine speeds and loads demonstrate that the proposed method surpasses conventional threshold-based indices and recent deep learning baselines, achieving superior accuracy, robustness, and computational efficiency suitable for real-time combustion monitoring and control-oriented applications. To ensure the reliability of the reported results, all metrics were averaged over three independent runs to mitigate the impact of network initialization and sampling randomness.


  • A Deep Reinforcement Learning Method for Autonomous Driving Integrating Multi-Modal Fusion

    by Chengcheng Xu, Haiyan Zhao, Xinghao Lu, Kang Sun, Bingzhao Gao, Hong Chen

    Due to the input of high-dimensional data in end-to-end autonomous driving, if the feature extraction network is trained online from scratch, it will lead to difficulties in learning and decision-making. To solve this challenging problem, this paper proposes a predefined multi-modal image method to improve decision-making accuracy and a shared framework to quickly train all networks. The inputs are composed of various observations including bird’s-eye view, state, and front view, rather than traditional sensor raw data. They can be filtered out ineffectual features based on human prior knowledge to avoid over-exploration with invalid changes, such as light, trees, mountains, etc. Combined with the processed representations, predefined images are formed by embedding the trajectory, forbidden line, etc. The method can assist agents to capture the coupling relationship between the actions and states, allowing networks to positively adjust weights to quickly obtain expected rewards. Considering that embedding a feature extraction network into reinforcement learning networks will lead to repeated cumulative calculations, a shared framework is proposed to independently compress features while jointly participates in online training to reduce computing costs. This module dynamically combines the policy and critic to update its weights, which overcomes decision-making problems caused by inaccurate the latent feature sequence during pre-training and fine-tuning methods. Finally, the interactive environment is constructed on a realistic driving simulator CARLA, and the fusion of different modal states is explored. The results showed that the multi-modal fusion can explore to earn maximum rewards, and the proposed methods are effective for the training.


  • Decision-Making Method Combining Path-Constraint-Based Action Compensation for Autonomous Vehicles

    by Haiyan Zhao, Jingdi Cao, Xinghao Lu, Chengcheng Xu, Hong Chen

    In order to improve the performance and reliability of the decision-making process for autonomous vehicles, a decision-making method using deep reinforcement learning (DRL) and combining path constraints-based action compensation is proposed in this paper. In the decision-making method, an action compensation module based on path constraints is developed, which enhances the reliability of the decision of vehicle and ensures that the final output actions meet safety and comfort requirements. Additionally, the traditional reinforcement learning task is divided into three subtasks, and the switching between the subtasks is realized by the scene recognition module. It reduces the complexity of the task and improves the training efficiency and sample utilization. Moreover, rewards and penalties are designed within path constraints, incorporating the distance from the lane center and the heading deviation angle. This can provide efficient and timely feedback for the vehicle during environmental interactions, accelerating the training process and improving decision-making performance. Finally, experiments are conducted using the CARLA simulation platform. The results show that the proposed method not only enables reliable decision-making but also improves the performance of vehicle, ensuring smooth control of the vehicle through the driving task, which verifies the effectiveness of the proposed method.


  • Multisensor Decision-Level Fusion Network Based on Attention Mechanism for Object Detection

    by Chengcheng Xu, Haiyan Zhao, Hongbin Xie, Bingzhao Gao

    To solve the problem of low accuracy caused by threshold constraints in traditional decision-level fusion methods, this paper proposes a deep learning method based on attention mechanism to fuse the three-dimensional information of sensors. The proposed model based on attention mechanism (AFnet) can improve the accuracy of the detection system without relying on traditional constraints. The AFnet model decouples the correlation between the data by the encoder, and fully utilizes the nonlinear fitting capability of deep learning. The adaptive fusion can be realized under data scale and result bias, which effectively solves the problem caused by traditional methods in the case of vehicle occlusion and overlap. The depth information and object detection networks are combined by embedding, which ensures that cameras can achieve spatial detection of vehicles and overcome the limitations of two-dimensional plane. The redefined clustering method takes into account the spatial position and velocity attribute, which can effectively distinguish high-density overlapping point clouds. Finally, experimental results in NuScenes and Carla show that the proposed fusion method does not rely on traditional rule constraints, and improves the accuracy of object detection. The fusion model of AFnet presents the state of the art on fusion matching accuracy of 99.11%.