Diffusion models have swiftly taken the lead in generative modeling, establishing unprecedented standards for producing high-quality, varied outputs. Unlike Generative Adversarial Networks (GANs)—once considered the gold standard in this realm—diffusion models bring several unique benefits to the table. They are renowned for generating outputs that more accurately reflect the complexity of real-world data, showcase a wider array of diversity, and are based on a training approach that is comparatively more straightforward and stable. This survey aims to offer an exhaustive overview of both the theoretical underpinnings and practical achievements of diffusion models. We explore and outline three core approaches to diffusion modeling: denoising diffusion probabilistic models, score-based generative models, and stochastic differential equations. Subsequently, we delineate the algorithmic enhancements of diffusion models across several pivotal areas. A notable aspect of this review is an in-depth analysis of leading generative models, examining how diffusion models relate to and evolve from previous generative methodologies, offering critical insights into their synergy. A comparative analysis of the merits and limitations of different generative models is a vital component of our discussion. Moreover, we highlight the applications of diffusion models across computer vision, multi-modal generation, and beyond, culminating in significant conclusions and suggesting promising avenues for future investigation.
9 May 2022 in IEEE Transactions on Computer-Aided Design of Integrated Circuits and Systems
Object segmentation is a per-pixel label prediction task which targets at providing context analysis for autonomous driving. Moving object segmentation (MOS) serves as a subbranch of object segmentation, targeting at separating the surrounding objects into binary options: dynamic and static. MOS is vital for safety-critical task in autonomous driving because dynamic objects are often true potential threat to self-driving car comparing to static ones. Current methods typically address the MOS problem as a category feature to label mapping task, which is not rational in reality. For example, a parking car should be considered as static instead of moving object category. There is little systematic theory to differentiate object moving characteristics from non-moving characteristics in MOS. Furthermore, restricted by limit resource in embedded system, MOS is often in an off-line manner due to huge computational requirement. An on-line and low computational cost MOS is an urgent demand for practical safety-critical mission which takes immediate reaction as compulsory. In this paper, we propose EmPointMovseg, an efficient and practical 3D LiDAR MOS solution for autonomous driving. Leveraging the power of well-adapted autoregressive system identification (AR-SI) theory, EmPointMovseg theoretically explains moving-object feature in large scale 3D LiDAR semantic segmentation. An end-to-end sparse tensor based CNN which balances segmentation accuracy and on-line process ability is proposed. We construct our experiment on both representative dataset benchmarks and practical embedded systems. The evaluation result shows the effectiveness and accuracy of our proposed solution, conquering the bottleneck in the on-line large-scale 3D LiDAR semantic segmentation.
9 April 2025 in ACM Transactions on Software Engineering and Methodology
Testing Autonomous Driving Systems (ADS) is critical for validating their safety in operational environments. High-fidelity simulators enable the testing of ADS through virtual driving scenarios, especially those that are hazardous to replicate in real-world settings. However, existing testing approaches suffer from inadequate coverage of real-world traffic situations due to over-simplified modeling of vehicle movements (e.g., insufficient diversity in driving styles), resulting in undetected critical ADS failures. In this paper, we propose a testing framework to discover diverse failures of ADS in driving scenarios that embody real-world traffic complexity. The framework leverages advanced traffic simulation methods to encode vehicle movements and generates realistic yet safety-critical driving scenarios for ADS by mutating vehicle movements. To efficiently explore driving scenarios that pose different challenges for ADS and expose diverse ADS failures, this framework further leverages a dynamic prioritization mechanism that prioritizes vehicle movements likely to trigger unique ADS behaviors. Specifically, we propose a method to estimate the possibility based on encoded vehicle movements. We implement this framework and evaluate it with three representative ADS from the famous CARLA leaderboard. Empirical evaluation demonstrates that the proposed approach discovers more unique failures of ADS than existing testing frameworks.
This review systematically examines major research advances in AUV path planning over recent years, covering several mainstream methodologies: sample-based path planning (e.g., PRM and RRT along with their asymptotically optimal variants, suitable for high-dimensional space exploration), graph-search-based path planning (e.g., A-series and D-series algorithms, achieving global optimization and dynamic replanning through environmental modeling), optimization-based approaches (including artificial potential field (APF), nonlinear programming (NLP), and model predictive control (MPC), designed to satisfy stringent dynamic constraints on AUV motion), swarm intelligence-based planning methods (such as genetic algorithms and ant colony optimization), and learning-based intelligent methods (such as deep reinforcement learning (DRL) for real-time decision-making in unknown and dynamic environments). Through in-depth analysis of these methods’ principles, improvement strategies, and AUV path planning contexts, this review highlights current research trends toward hybrid cooperative planning, dynamic environmental adaptability, and high-precision trajectory optimization. Finally, the paper outlines future directions for AUV path planning, emphasizing multi-AUV collaboration and higher-level intelligent decision-making as key research priorities.
19 January 2024 in ACM Transactions on Embedded Computing Systems
Autonomous driving emphasizes precise multi-sensor fusion positioning on limit resource embedded systems. LiDAR-centered sensor fusion system serves as a mainstream navigation system due to its insensitivity to illumination and viewpoint change. However, these types of systems suffer from handling large-scale sequential LiDAR data using limited resources on board, leading LiDAR-centralized sensor fusion unpractical. As a result, hand-crafted features such as plane and edge are leveraged in majority mainstream positioning methods to alleviate this unsatisfaction, triggering a new cornerstone in LiDAR Inertial sensor fusion. However, such super light weight feature extraction, although it achieves real-time constraint in LiDAR-centered sensor fusion, encounters severe vulnerability under high speed rotational or translational perturbation. In this paper, we propose a sparse tensor based LiDAR Inertial fusion method for autonomous driving embedded system. Leveraging the power of sparse tensor, the global geometrical feature is fetched so that the point cloud sparsity defect is alleviated. Inertial sensor is deployed to conquer the time-consuming step caused by the coarse level point-wise inlier matching. We construct our experiments on both representative dataset benchmarks and realistic scenes. The evaluation results show the robustness and accuracy of our proposed solution compared to classical methods.
24 February 2025 in IEEE Transactions on Instrumentation and Measurement
State estimation encounters significant hurdles in scale ambiguity, both when assimilating data from scale-uninformed sources such as Structure from Motion (SfM) and when handling normalized point clouds, each scenario demanding robust solutions to achieve consistent scale and accurate estimation. Addressing this critical issue, we propose the Scaled Quadratic Pose Estimation Problem (sQPEP), a novel unified framework designed to enhance scale estimation in various state estimation algorithms. Our framework not only establishes a globally optimal solution strategy for the precise estimation of pose and scale factors but also systematically categorizes a broad spectrum of pose estimation challenges. This is crucial for advancing our theoretical understanding and the practical application of these solutions. The sQPEP framework consolidates a range of scale and pose estimation challenges into a unified theoretical paradigm, thereby refining the methodology for these estimations. By applying algebraic techniques, we have effectively bifurcated the problem into two distinct categories. Subsequently, we have deduced globally optimal solutions and unveiled two robust solvers. These solvers are proficient in generating 80 and 81 solutions for their respective problem classes, featuring elimination template dimensions of 664×744 and 521×602. Our method’s efficacy has been rigorously confirmed through experimental validation, which demonstrates its consistent performance in degenerate conditions and its superior noise immunity. These results bolster the framework’s applicability to intricate scenarios encountered in real-world settings.