arXiv subjects

Junyue Jiang

Publications and source records attributed to Junyue Jiang.

5 recordsLinked to original sources

arXiv2025

Towards Accurate and Efficient 3D Object Detection for Autonomous Driving: A Mixture of Experts Computing System on Edge

This paper presents Edge-based Mixture of Experts (MoE) Collaborative Computing (EMC2), an optimal computing system designed for autonomous vehicles (AVs) that simultaneously achieves low-latency and high-accuracy 3D object detection. Unlike conventional approaches, EMC2 incorporates a scenario-aware MoE architecture specifically optimized for edge platforms. By effectively fusing LiDAR and camera data, the system leverages the complementary strengths of sparse 3D point clouds and dense 2D images to generate robust multimodal representations. To enable this, EMC2 employs an adaptive multimodal data bridge that performs multi-scale preprocessing on sensor inputs, followed by a scenario-aware routing mechanism that dynamically dispatches features to dedicated expert models based on object visibility and distance. In addition, EMC2 integrates joint hardware-software optimizations, including hardware resource utilization optimization and computational graph simplification, to ensure efficient and real-time inference on resource-constrained edge devices. Experiments on open-source benchmarks clearly show the EMC2 advancements as an end-to-end system. On the KITTI dataset, it achieves an average accuracy improvement of 3.58% and a 159.06% inference speedup compared to 15 baseline methods on Jetson platforms, with similar performance gains on the nuScenes dataset, highlighting its capability to advance reliable, real-time 3D object detection tasks for AVs. The official implementation is available at https://github.com/LinshenLiu622/EMC2.

cs.CV↗

arXiv2024

Trajectory Tracking Using Frenet Coordinates with Deep Deterministic Policy Gradient

This paper studies the application of the DDPG algorithm in trajectory-tracking tasks and proposes a trajectorytracking control method combined with Frenet coordinate system. By converting the vehicle's position and velocity information from the Cartesian coordinate system to Frenet coordinate system, this method can more accurately describe the vehicle's deviation and travel distance relative to the center line of the road. The DDPG algorithm adopts the Actor-Critic framework, uses deep neural networks for strategy and value evaluation, and combines the experience replay mechanism and target network to improve the algorithm's stability and data utilization efficiency. Experimental results show that the DDPG algorithm based on Frenet coordinate system performs well in trajectory-tracking tasks in complex environments, achieves high-precision and stable path tracking, and demonstrates its application potential in autonomous driving and intelligent transportation systems. Keywords- DDPG; path tracking; robot navigation

cs.RO↗

arXiv2024

Towards Explainable Traffic Flow Prediction with Large Language Models

Traffic forecasting is crucial for intelligent transportation systems. It has experienced significant advancements thanks to the power of deep learning in capturing latent patterns of traffic data. However, recent deep-learning architectures require intricate model designs and lack an intuitive understanding of the mapping from input data to predicted results. Achieving both accuracy and explainability in traffic prediction models remains a challenge due to the complexity of traffic data and the inherent opacity of deep learning models. To tackle these challenges, we propose a Traffic flow Prediction model based on Large Language Models (LLMs) to generate explainable traffic predictions, named xTP-LLM. By transferring multi-modal traffic data into natural language descriptions, xTP-LLM captures complex time-series patterns and external factors from comprehensive traffic data. The LLM framework is fine-tuned using language-based instructions to align with spatial-temporal traffic flow data. Empirically, xTP-LLM shows competitive accuracy compared with deep learning baselines, while providing an intuitive and reliable explanation for predictions. This paper contributes to advancing explainable traffic prediction models and lays a foundation for future exploration of LLM applications in transportation. To the best of our knowledge, this is the first study to use LLM for explainable prediction of traffic flows.

cs.LG↗

arXiv2023

Conditional Flood Fill Method in Logic Synthesis

In the field of Electronic Design Automation (EDA), logic synthesis plays a pivotal role in optimizing hardware resources. Traditional logic synthesis algorithms, such as the Quine-McCluskey method, face challenges in scalability and efficiency, particularly for higher-dimension problems. This paper introduces a novel heuristic algorithm based on Conditional Flood Fill Method aimed at addressing these limitations. Our method employs count-based adjacent element handling and introduces nine new theorems to guide the logic synthesis process. Experimental results validate the efficacy of our approach, showing significant improvements in computational efficiency and scalability compared to existing algorithms. The algorithm holds potential for future advancements in circuit development and Boolean function optimization.

cs.AR↗

arXiv2023

FiND: Few-shot three-dimensional image-free confocal focusing on point-like emitters

Confocal fluorescence microscopy is widely applied for the study of point-like emitters such as biomolecules, material defects, and quantum light sources. Confocal techniques offer increased optical resolution, dramatic fluorescence background rejection and sub-nanometer localization, useful in super-resolution imaging of fluorescent biomarkers, single-molecule tracking, or the characterization of quantum emitters. However, rapid, noise-robust automated 3D focusing on point-like emitters has been missing for confocal microscopes. Here, we introduce FiND (Focusing in Noisy Domain), an imaging-free, non-trained 3D focusing framework that requires no hardware add-ons or modifications. FiND achieves focusing for signal-to-noise ratios down to 1, with a few-shot operation for signal-to-noise ratios above 5. FiND enables unsupervised, large-scale focusing on a heterogeneous set of quantum emitters. Additionally, we demonstrate the potential of FiND for real-time 3D tracking by following the drift trajectory of a single NV center indefinitely with a positional precision of < 10 nm. Our results show that FiND is a useful focusing framework for the scalable analysis of point-like emitters in biology, material science, and quantum optics.

physics.optics↗