Yuxiang Sun

Zhenhua Xu, Yuxuan Liu, Yuxiang Sun et al. · gatech

The road network graph is a critical component for downstream tasks in autonomous driving, such as global route planning and navigation. In the past years, road network graphs are usually annotated by human experts manually, which is time-consuming and labor-intensive. To annotate road network graphs effectively and efficiently, automatic algorithms for road network graph detection are demanded. Most existing methods either adopt a post-processing step on semantic segmentation maps to produce road network graphs, or propose graph-based algorithms to directly predict the graphs. However, these works suffer from hard-coded algorithms and inferior performance. To enhance the previous state-of-the-art (SOTA) method RNGDet, we add an instance segmentation head to better supervise the training, and enable the network to leverage multi-scale features of the backbone. Since the new proposed approach is improved from RNGDet, we name it RNGDet++. Experimental results show that our RNGDet++ outperforms baseline methods in terms of almost all evaluation metrics on two large-scale public datasets. Our code and supplementary materials are available at \url{https://tonyxuqaq.github.io/projects/RNGDetPlusPlus/}.

Zhenhua Xu, Yuxuan Liu, Yuxiang Sun et al. · gatech

With the fast development of autonomous driving technologies, there is an increasing demand for high-definition (HD) maps, which provide reliable and robust prior information about the static part of the traffic environments. As one of the important elements in HD maps, road lane centerline is critical for downstream tasks, such as prediction and planning. Manually annotating centerlines for road lanes in HD maps is labor-intensive, expensive and inefficient, severely restricting the wide applications of autonomous driving systems. Previous work seldom explores the lane centerline detection problem due to the complicated topology and severe overlapping issues of lane centerlines. In this paper, we propose a novel method named CenterLineDet to detect lane centerlines for automatic HD map generation. Our CenterLineDet is trained by imitation learning and can effectively detect the graph of centerlines with vehicle-mounted sensors (i.e., six cameras and one LiDAR) through iterations. Due to the use of the DETR-like transformer network, CenterLineDet can handle complicated graph topology, such as lane intersections. The proposed approach is evaluated on the large-scale public dataset NuScenes. The superiority of our CenterLineDet is demonstrated by the comparative results. Our code, supplementary materials, and video demonstrations are available at \href{https://tonyxuqaq.github.io/projects/CenterLineDet/}{https://tonyxuqaq.github.io/projects/CenterLineDet/}.

Zhen Feng, Yuchao Feng, Yanning Guo et al.

Segmentation of drivable roads and negative obstacles is critical to the safe driving of autonomous vehicles. Currently, many multi-modal fusion methods have been proposed to improve segmentation accuracy, such as fusing RGB and depth images. However, we find that when fusing two modals of data with untrustworthy features, the performance of multi-modal networks could be degraded, even lower than those using a single modality. In this paper, the untrustworthy features refer to those extracted from regions (e.g., far objects that are beyond the depth measurement range) with invalid depth data (i.e., 0 pixel value) in depth images. The untrustworthy features can confuse the segmentation results, and hence lead to inferior results. To provide a solution to this issue, we propose the Adaptive-Mask Fusion Network (AMFNet) by introducing adaptive-weight masks in the fusion module to fuse features from RGB and depth images with inconsistency. In addition, we release a large-scale RGB-depth dataset with manually-labeled ground truth based on the NPO dataset for drivable roads and negative obstacles segmentation. Extensive experimental results demonstrate that our network achieves state-of-the-art performance compared with other networks. Our code and dataset are available at: https://github.com/lab-sun/AMFNet.

Jianxin Huang, Jiahang Li, Ning Jia et al.

Task-specific data-fusion networks have marked considerable achievements in urban scene parsing. Among these networks, our recently proposed RoadFormer successfully extracts heterogeneous features from RGB images and surface normal maps and fuses these features through attention mechanisms, demonstrating compelling efficacy in RGB-Normal road scene parsing. However, its performance significantly deteriorates when handling other types/sources of data or performing more universal, all-category scene parsing tasks. To overcome these limitations, this study introduces RoadFormer+, an efficient, robust, and adaptable model capable of effectively fusing RGB-X data, where ``X'', represents additional types/modalities of data such as depth, thermal, surface normal, and polarization. Specifically, we propose a novel hybrid feature decoupling encoder to extract heterogeneous features and decouple them into global and local components. These decoupled features are then fused through a dual-branch multi-scale heterogeneous feature fusion block, which employs parallel Transformer attentions and convolutional neural network modules to merge multi-scale features across different scales and receptive fields. The fused features are subsequently fed into a decoder to generate the final semantic predictions. Notably, our proposed RoadFormer+ ranks first on the KITTI Road benchmark and achieves state-of-the-art performance in mean intersection over union on the Cityscapes, MFNet, FMB, and ZJU datasets. Moreover, it reduces the number of learnable parameters by 65\% compared to RoadFormer. Our source code will be publicly available at mias.group/RoadFormerPlus.

Xinyu Gao, Ziyi Yang, Yunlu Zhao et al.

A variety of Neural Radiance Fields (NeRF) methods have recently achieved remarkable success in high render speed. However, current accelerating methods are specialized and incompatible with various implicit methods, preventing real-time composition over various types of NeRF works. Because NeRF relies on sampling along rays, it is possible to provide general guidance for acceleration. To that end, we propose a general implicit pipeline for composing NeRF objects quickly. Our method enables the casting of dynamic shadows within or between objects using analytical light sources while allowing multiple NeRF objects to be seamlessly placed and rendered together with any arbitrary rigid transformations. Mainly, our work introduces a new surface representation known as Neural Depth Fields (NeDF) that quickly determines the spatial relationship between objects by allowing direct intersection computation between rays and implicit surfaces. It leverages an intersection neural network to query NeRF for acceleration instead of depending on an explicit spatial structure.Our proposed method is the first to enable both the progressive and interactive composition of NeRF objects. Additionally, it also serves as a previewing plugin for a range of existing NeRF works.

Lincan Li, Bolin Shen, Chenxi Zhao et al.

Graph-structured data, which captures non-Euclidean relationships and interactions between entities, is growing in scale and complexity. As a result, training state-of-the-art graph machine learning (GML) models have become increasingly resource-intensive, turning these models and data into invaluable Intellectual Property (IP). To address the resource-intensive nature of model training, graph-based Machine-Learning-as-a-Service (GMLaaS) has emerged as an efficient solution by leveraging third-party cloud services for model development and management. However, deploying such models in GMLaaS also exposes them to potential threats from attackers. Specifically, while the APIs within a GMLaaS system provide interfaces for users to query the model and receive outputs, they also allow attackers to exploit and steal model functionalities or sensitive training data, posing severe threats to the safety of these GML models and the underlying graph data. To address these challenges, this survey systematically introduces the first taxonomy of threats and defenses at the level of both GML model and graph-structured data. Such a tailored taxonomy facilitates an in-depth understanding of GML IP protection. Furthermore, we present a systematic evaluation framework to assess the effectiveness of IP protection methods, introduce a curated set of benchmark datasets across various domains, and discuss their application scopes and future challenges. Finally, we establish an open-sourced versatile library named PyGIP, which evaluates various attack and defense techniques in GMLaaS scenarios and facilitates the implementation of existing benchmark methods. The library resource can be accessed at: https://labrai.github.io/PyGIP. We believe this survey will play a fundamental role in intellectual property protection for GML and provide practical recipes for the GML community.

Huaiyang Huang, Haoyang Ye, Jianhao Jiao et al.

Visual Localization is an essential component in autonomous navigation. Existing approaches are either based on the visual structure from SLAM/SfM or the geometric structure from dense mapping. To take the advantages of both, in this work, we present a complete visual inertial localization system based on a hybrid map representation to reduce the computational cost and increase the positioning accuracy. Specially, we propose two modules for data association and batch optimization, respectively. To this end, we develop an efficient data association module to associate map components with local features, which takes only $2$ms to generate temporal landmarks. For batch optimization, instead of using visual factors, we develop a module to estimate a pose prior from the instant localization results to constrain poses. The experimental results on the EuRoC MAV dataset demonstrate a competitive performance compared to the state of the arts. Specially, our system achieves an average position error in 1.7 cm with 100% recall. The timings show that the proposed modules reduce the computational cost by 20-30%. We will make our implementation open source at http://github.com/hyhuang1995/gmmloc.

Xudong Lv, Yuxiang Sun, Shuo Wang et al.

Optical neural networks are emerging as powerful machine learning and information processing tools because of their potential advantages in speed and energy efficiency. The training methods of these physical models, however, remain underexplored compared to their digital counterparts and are leading to suboptimal performance. This paper reports a pre-training-driven approach that leads to snapshot image denoising with substantially improved quality. We demonstrated effective free-space optical denoising by a diffractive network optimized by a two-step process including (1) pre-training using a massive dataset of 3.45 million diverse but simple images and (2) fine-tuning with the corresponding task-specific datasets. Compared to conventional Fourier-domain filtering and directly trained diffractive networks, such a transfer learning process exhibited prominent advantages for denoising images degraded by severe noise, peak signal-to-noise ratio (PSNR) below 8 dB, while preserving fine image features and improving the PSNR to above 18 dB. Importantly, the same pre-trained optical network could be consistently fine-tuned to process degraded images from highly diverse styles ranging from handwritten digits (MNIST) and chest X-rays (ChestMNIST) to CIFAR-10 images and human faces (CelebA). We further demonstrated the critical role of our optical denoisers in vision-based applications, including face detection, plate recognition, and localization of UAVs in noisy conditions.

Mengshi Chen, Yuxiang Sun, Tengchao Li et al.

Data preparation is a critical step in enhancing the usability of tabular data and thus boosts downstream data-driven tasks. Traditional methods often face challenges in capturing the intricate relationships within tables and adapting to the tasks involved. Recent advances in Language Models (LMs), especially in Large Language Models (LLMs), offer new opportunities to automate and support tabular data preparation. However, why LMs suit tabular data preparation (i.e., how their capabilities match task demands) and how to use them effectively across phases still remain to be systematically explored. In this survey, we systematically analyze the role of LMs in enhancing tabular data preparation processes, focusing on four core phases: data acquisition, integration, cleaning, and transformation. For each phase, we present an integrated analysis of how LMs can be combined with other components for different preparation tasks, highlight key advancements, and outline prospective pipelines.

Guoyan Wang, Yanyan Huang, Chunlin Chen et al.

Automated operation in cross-platform strategy games demands agents with robust generalization across diverse user interfaces and dynamic battlefield conditions. While vision-language models (VLMs) have shown considerable promise in multimodal reasoning, their application to complex human-computer interaction scenarios--such as strategy gaming--remains largely unexplored. Here, we introduce Yanyun-3, a general-purpose agent framework that, for the first time, enables autonomous cross-platform operation across three heterogeneous strategy game environments. By integrating the vision-language reasoning of Qwen2.5-VL with the precise execution capabilities of UI-TARS, Yanyun-3 successfully performs core tasks including target localization, combat resource allocation, and area control. Through systematic ablation studies, we evaluate the effects of various multimodal data combinations--static images, multi-image sequences, and videos--and propose the concept of combination granularity to differentiate between intra-sample fusion and inter-sample mixing strategies. We find that a hybrid strategy, which fuses multi-image and video data while mixing in static images (MV+S), substantially outperforms full fusion: it reduces inference time by 63% and boosts the BLEU-4 score by a factor of 12 (from 4.81% to 62.41%, approximately 12.98x). Operating via a closed-loop pipeline of screen capture, model inference, and action execution, the agent demonstrates strong real-time performance and cross-platform generalization. Beyond providing an efficient solution for strategy game automation, our work establishes a general paradigm for enhancing VLM performance through structured multimodal data organization, offering new insights into the interplay between static perception and dynamic reasoning in embodied intelligence.

Xuhua Zhao, Yuxuan Xie, Caihua Chen et al.

Recent advances in mathematical reasoning and the long-term planning capabilities of large language models (LLMs) have precipitated the development of agents, which are being increasingly leveraged in business operations processes. Decision models to optimize inventory levels are one of the core elements of operations management. However, the capabilities of the LLM agent in making inventory decisions in uncertain contexts, as well as the decision-making biases (e.g. framing effect, etc.) of the agent, remain largely unexplored. This prompts concerns regarding the capacity of LLM agents to effectively address real-world problems, as well as the potential implications of biases that may be present. To address this gap, we introduce AIM-Bench, a novel benchmark designed to assess the decision-making behaviour of LLM agents in uncertain supply chain management scenarios through a diverse series of inventory replenishment experiments. Our results reveal that different LLMs typically exhibit varying degrees of decision bias that are similar to those observed in human beings. In addition, we explored strategies to mitigate the pull-to-centre effect and the bullwhip effect, namely cognitive reflection and implementation of information sharing. These findings underscore the need for careful consideration of the potential biases in deploying LLMs in Inventory decision-making scenarios. We hope that these insights will pave the way for mitigating human decision bias and developing human-centred decision support systems for supply chains.

Bowen Zhang, Ying Chen, Long Bai et al.

Foundation models have become prominent in computer vision, achieving notable success in various tasks. However, their effectiveness largely depends on pre-training with extensive datasets. Applying foundation models directly to small datasets of capsule endoscopy images from scratch is challenging. Pre-training on broad, general vision datasets is crucial for successfully fine-tuning our model for specific tasks. In this work, we introduce a simplified approach called Adapt foundation models with a low-rank adaptation (LoRA) technique for easier customization. Our method, inspired by the DINOv2 foundation model, applies low-rank adaptation learning to tailor foundation models for capsule endoscopy diagnosis effectively. Unlike traditional fine-tuning methods, our strategy includes LoRA layers designed to absorb specific surgical domain knowledge. During the training process, we keep the main model (the backbone encoder) fixed and focus on optimizing the LoRA layers and the disease classification component. We tested our method on two publicly available datasets for capsule endoscopy disease classification. The results were impressive, with our model achieving 97.75% accuracy on the Kvasir-Capsule dataset and 98.81% on the Kvasirv2 dataset. Our solution demonstrates that foundation models can be adeptly adapted for capsule endoscopy diagnosis, highlighting that mere reliance on straightforward fine-tuning or pre-trained models from general computer vision tasks is inadequate for such specific applications.

Zhenhua Xu, Yuxuan Liu, Lu Gan et al.

Road network graphs provide critical information for autonomous-vehicle applications, such as drivable areas that can be used for motion planning algorithms. To find road network graphs, manually annotation is usually inefficient and labor-intensive. Automatically detecting road network graphs could alleviate this issue, but existing works still have some limitations. For example, segmentation-based approaches could not ensure satisfactory topology correctness, and graph-based approaches could not present precise enough detection results. To provide a solution to these problems, we propose a novel approach based on transformer and imitation learning in this paper. In view of that high-resolution aerial images could be easily accessed all over the world nowadays, we make use of aerial images in our approach. Taken as input an aerial image, our approach iteratively generates road network graphs vertex-by-vertex. Our approach can handle complicated intersection points with various numbers of incident road segments. We evaluate our approach on a publicly available dataset. The superiority of our approach is demonstrated through the comparative experiments. Our work is accompanied with a demonstration video which is available at \url{https://tonyxuqaq.github.io/projects/RNGDet/}.

M. Usman Maqbool Bhutta, Yuxiang Sun, Darwin Lau et al.

Deep learning-based image retrieval techniques for the loop closure detection demonstrate satisfactory performance. However, it is still challenging to achieve high-level performance based on previously trained models in different geographical regions. This paper addresses the problem of their deployment with simultaneous localization and mapping (SLAM) systems in the new environment. The general baseline approach uses additional information, such as GPS, sequential keyframes tracking, and re-training the whole environment to enhance the recall rate. We propose a novel approach for improving image retrieval based on previously trained models. We present an intelligent method, MAQBOOL, to amplify the power of pre-trained models for better image recall and its application to real-time multiagent SLAM systems. We achieve comparable image retrieval results at a low descriptor dimension (512-D), compared to the high descriptor dimension (4096-D) of state-of-the-art methods. We use spatial information to improve the recall rate in image retrieval on pre-trained models.

Zhenhua Xu, Yuxuan Liu, Lu Gan et al.

High-Definition (HD) maps can provide precise geometric and semantic information of static traffic environments for autonomous driving. Road-boundary is one of the most important information contained in HD maps since it distinguishes between road areas and off-road areas, which can guide vehicles to drive within road areas. But it is labor-intensive to annotate road boundaries for HD maps at the city scale. To enable automatic HD map annotation, current work uses semantic segmentation or iterative graph growing for road-boundary detection. However, the former could not ensure topological correctness since it works at the pixel level, while the latter suffers from inefficiency and drifting issues. To provide a solution to the aforementioned problems, in this letter, we propose a novel system termed csBoundary to automatically detect road boundaries at the city scale for HD map annotation. Our network takes as input an aerial image patch, and directly infers the continuous road-boundary graph (i.e., vertices and edges) from this image. To generate the city-scale road-boundary graph, we stitch the obtained graphs from all the image patches. Our csBoundary is evaluated and compared on a public benchmark dataset. The results demonstrate our superiority. The accompanied demonstration video is available at our project page \url{https://sites.google.com/view/csboundary/}.

Yuxiang Sun, Bo Yuan, Yufan Xue et al.

Researchers are increasingly focusing on intelligent games as a hot research area.The article proposes an algorithm that combines the multi-attribute management and reinforcement learning methods, and that combined their effect on wargaming, it solves the problem of the agent's low rate of winning against specific rules and its inability to quickly converge during intelligent wargame training.At the same time, this paper studied a multi-attribute decision making and reinforcement learning algorithm in a wargame simulation environment, and obtained data on red and blue conflict.Calculate the weight of each attribute based on the intuitionistic fuzzy number weight calculations. Then determine the threat posed by each opponent's chess pieces.Using the red side reinforcement learning reward function, the AC framework is trained on the reward function, and an algorithm combining multi-attribute decision-making with reinforcement learning is obtained. A simulation experiment confirms that the algorithm of multi-attribute decision-making combined with reinforcement learning presented in this paper is significantly more intelligent than the pure reinforcement learning algorithm.By resolving the shortcomings of the agent's neural network, coupled with sparse rewards in large-map combat games, this robust algorithm effectively reduces the difficulties of convergence. It is also the first time in this field that an algorithm design for intelligent wargaming combines multi-attribute decision making with reinforcement learning.Attempt interdisciplinary cross-innovation in the academic field, like designing intelligent wargames and improving reinforcement learning algorithms.

Peide Cai, Hengli Wang, Yuxiang Sun et al.

Autonomous driving in multi-agent dynamic traffic scenarios is challenging: the behaviors of road users are uncertain and are hard to model explicitly, and the ego-vehicle should apply complicated negotiation skills with them, such as yielding, merging and taking turns, to achieve both safe and efficient driving in various settings. Traditional planning methods are largely rule-based and scale poorly in these complex dynamic scenarios, often leading to reactive or even overly conservative behaviors. Therefore, they require tedious human efforts to maintain workability. Recently, deep learning-based methods have shown promising results with better generalization capability but less hand engineering efforts. However, they are either implemented with supervised imitation learning (IL), which suffers from dataset bias and distribution mismatch issues, or are trained with deep reinforcement learning (DRL) but focus on one specific traffic scenario. In this work, we propose DQ-GAT to achieve scalable and proactive autonomous driving, where graph attention-based networks are used to implicitly model interactions, and deep Q-learning is employed to train the network end-to-end in an unsupervised manner. Extensive experiments in a high-fidelity driving simulator show that our method achieves higher success rates than previous learning-based methods and a traditional rule-based method, and better trades off safety and efficiency in both seen and unseen scenarios. Moreover, qualitative results on a trajectory dataset indicate that our learned policy can be transferred to the real world for practical applications with real-time speeds. Demonstration videos are available at https://caipeide.github.io/dq-gat/.

Huaiyang Huang, Yuxiang Sun, Jin Wu et al.

Multiview registration is used to estimate Rigid Body Transformations (RBTs) from multiple frames and reconstruct a scene with corresponding scans. Despite the success of pairwise registration and pose synchronization, the concept of Bundle Adjustment (BA) has been proven to better maintain global consistency. So in this work, we make the multiview point-cloud registration more tractable from a different perspective in resolving range-based BA. Based on this analysis, we propose an objective function that takes both measurement noises and computational cost into account. For the feature parameter update, instead of calculating the global distribution parameters from the raw measurements, we aggregate the local distributions upon the pose update at each iteration. The computational cost of feature update is then only dependent on the number of scans. Finally, we develop a multiview registration system using voxel-based quantization that can be applied in real-world scenarios. The experimental results demonstrate our superiority over the baselines in terms of both accuracy and speed. Moreover, the results also show that our average positioning errors achieve the centimeter level.

Huaiyang Huang, Haoyang Ye, Yuxiang Sun et al.

Traditional approaches for Visual Simultaneous Localization and Mapping (VSLAM) rely on low-level vision information for state estimation, such as handcrafted local features or the image gradient. While significant progress has been made through this track, under more challenging configuration for monocular VSLAM, e.g., varying illumination, the performance of state-of-the-art systems generally degrades. As a consequence, robustness and accuracy for monocular VSLAM are still widely concerned. This paper presents a monocular VSLAM system that fully exploits learnt features for better state estimation. The proposed system leverages both learnt local features and global embeddings at different modules of the system: direct camera pose estimation, inter-frame feature association, and loop closure detection. With a probabilistic explanation of keypoint prediction, we formulate the camera pose tracking in a direct manner and parameterize local features with uncertainty taken into account. To alleviate the quantization effect, we adapt the mapping module to generate 3D landmarks better to guarantee the system's robustness. Detecting temporal loop closure via deep global embeddings further improves the robustness and accuracy of the proposed system. The proposed system is extensively evaluated on public datasets (Tsukuba, EuRoC, and KITTI), and compared against the state-of-the-art methods. The competitive performance of camera pose estimation confirms the effectiveness of our method.

Zhenhua Xu, Yuxiang Sun, Lujia Wang et al.

Road curb detection is important for autonomous driving. It can be used to determine road boundaries to constrain vehicles on roads, so that potential accidents could be avoided. Most of the current methods detect road curbs online using vehicle-mounted sensors, such as cameras or 3-D Lidars. However, these methods usually suffer from severe occlusion issues. Especially in highly-dynamic traffic environments, most of the field of view is occupied by dynamic objects. To alleviate this issue, we detect road curbs offline using high-resolution aerial images in this paper. Moreover, the detected road curbs can be used to create high-definition (HD) maps for autonomous vehicles. Specifically, we first predict the pixel-wise segmentation map of road curbs, and then conduct a series of post-processing steps to extract the graph structure of road curbs. To tackle the disconnectivity issue in the segmentation maps, we propose an innovative connectivity-preserving loss (CP-loss) to improve the segmentation performance. The experimental results on a public dataset demonstrate the effectiveness of our proposed loss function. This paper is accompanied with a demonstration video and a supplementary document, which are available at \texttt{\url{https://sites.google.com/view/cp-loss}}.

Hengli Wang, Peide Cai, Rui Fan et al.

With the recent advancement of deep learning technology, data-driven approaches for autonomous car prediction and planning have achieved extraordinary performance. Nevertheless, most of these approaches follow a non-interactive prediction and planning paradigm, hypothesizing that a vehicle's behaviors do not affect others. The approaches based on such a non-interactive philosophy typically perform acceptably in sparse traffic scenarios but can easily fail in dense traffic scenarios. Therefore, we propose an end-to-end interactive neural motion planner (INMP) for autonomous driving in this paper. Given a set of past surrounding-view images and a high definition map, our INMP first generates a feature map in bird's-eye-view space, which is then processed to detect other agents and perform interactive prediction and planning jointly. Also, we adopt an optical flow distillation paradigm, which can effectively improve the network performance while still maintaining its real-time inference speed. Extensive experiments on the nuScenes dataset and in the closed-loop Carla simulation environment demonstrate the effectiveness and efficiency of our INMP for the detection, prediction, and planning tasks. Our project page is at sites.google.com/view/inmp-ofd.

Hengli Wang, Peide Cai, Yuxiang Sun et al.

Recently, deep-learning based approaches have achieved impressive performance for autonomous driving. However, end-to-end vision-based methods typically have limited interpretability, making the behaviors of the deep networks difficult to explain. Hence, their potential applications could be limited in practice. To address this problem, we propose an interpretable end-to-end vision-based motion planning approach for autonomous driving, referred to as IVMP. Given a set of past surrounding-view images, our IVMP first predicts future egocentric semantic maps in bird's-eye-view space, which are then employed to plan trajectories for self-driving vehicles. The predicted future semantic maps not only provide useful interpretable information, but also allow our motion planning module to handle objects with low probability, thus improving the safety of autonomous driving. Moreover, we also develop an optical flow distillation paradigm, which can effectively enhance the network while still maintaining its real-time performance. Extensive experiments on the nuScenes dataset and closed-loop simulation show that our IVMP significantly outperforms the state-of-the-art approaches in imitating human drivers with a much higher success rate. Our project page is available at https://sites.google.com/view/ivmp.

Zhenhua Xu, Yuxiang Sun, Ming Liu

Road-boundary detection is important for autonomous driving. It can be used to constrain autonomous vehicles running on road areas to ensure driving safety. Compared with online road-boundary detection using on-vehicle cameras/Lidars, offline detection using aerial images could alleviate the severe occlusion issue. Moreover, the offline detection results can be directly employed to annotate high-definition (HD) maps. In recent years, deep-learning technologies have been used in offline detection. But there still lacks a publicly available dataset for this task, which hinders the research progress in this area. So in this paper, we propose a new benchmark dataset, named \textit{Topo-boundary}, for offline topological road-boundary detection. The dataset contains 25,295 $1000\times1000$-sized 4-channel aerial images. Each image is provided with 8 training labels for different sub-tasks. We also design a new entropy-based metric for connectivity evaluation, which could better handle noises or outliers. We implement and evaluate 3 segmentation-based baselines and 5 graph-based baselines using the dataset. We also propose a new imitation-learning-based baseline which is enhanced from our previous work. The superiority of our enhancement is demonstrated from the comparison. The dataset and our-implemented code for the baselines are available at \texttt{\url{https://tonyxuqaq.github.io/Topo-boundary/}}.

Zhenhua Xu, Yuxiang Sun, Ming Liu

Detection of road curbs is an essential capability for autonomous driving. It can be used for autonomous vehicles to determine drivable areas on roads. Usually, road curbs are detected on-line using vehicle-mounted sensors, such as video cameras and 3-D Lidars. However, on-line detection using video cameras may suffer from challenging illumination conditions, and Lidar-based approaches may be difficult to detect far-away road curbs due to the sparsity issue of point clouds. In recent years, aerial images are becoming more and more worldwide available. We find that the visual appearances between road areas and off-road areas are usually different in aerial images, so we propose a novel solution to detect road curbs off-line using aerial images. The input to our method is an aerial image, and the output is directly a graph (i.e., vertices and edges) representing road curbs. To this end, we formulate the problem as an imitation learning problem, and design a novel network and an innovative training strategy to train an agent to iteratively find the road-curb graph. The experimental results on a public dataset confirm the effectiveness and superiority of our method. This work is accompanied with a demonstration video and a supplementary document at https://tonyxuqaq.github.io/iCurb/.

Hengli Wang, Yuxiang Sun, Rui Fan et al.

Path planning is a fundamental capability for autonomous navigation of robotic wheelchairs. With the impressive development of deep-learning technologies, imitation learning-based path planning approaches have achieved effective results in recent years. However, the disadvantages of these approaches are twofold: 1) they may need extensive time and labor to record expert demonstrations as training data; and 2) existing approaches could only receive high-level commands, such as turning left/right. These commands could be less sufficient for the navigation of mobile robots (e.g., robotic wheelchairs), which usually require exact poses of goals. We contribute a solution to this problem by proposing S2P2, a self-supervised goal-directed path planning approach. Specifically, we develop a pipeline to automatically generate planned path labels given as input RGB-D images and poses of goals. Then, we present a best-fit regression plane loss to train our data-driven path planning model based on the generated labels. Our S2P2 does not need pre-built maps, but it can be integrated into existing map-based navigation systems through our framework. Experimental results show that our S2P2 outperforms traditional path planning algorithms, and increases the robustness of existing map-based navigation systems. Our project page is available at https://sites.google.com/view/s2p2.

Hengli Wang, Rui Fan, Yuxiang Sun et al.

Joint detection of drivable areas and road anomalies is very important for mobile robots. Recently, many semantic segmentation approaches based on convolutional neural networks (CNNs) have been proposed for pixel-wise drivable area and road anomaly detection. In addition, some benchmark datasets, such as KITTI and Cityscapes, have been widely used. However, the existing benchmarks are mostly designed for self-driving cars. There lacks a benchmark for ground mobile robots, such as robotic wheelchairs. Therefore, in this paper, we first build a drivable area and road anomaly detection benchmark for ground mobile robots, evaluating the existing state-of-the-art single-modal and data-fusion semantic segmentation CNNs using six modalities of visual features. Furthermore, we propose a novel module, referred to as the dynamic fusion module (DFM), which can be easily deployed in existing data-fusion networks to fuse different types of visual features effectively and efficiently. The experimental results show that the transformed disparity image is the most informative visual feature and the proposed DFM-RTFNet outperforms the state-of-the-arts. Additionally, our DFM-RTFNet achieves competitive performance on the KITTI road benchmark. Our benchmark is publicly available at https://sites.google.com/view/gmrb.

Xiaodong Mei, Yuxiang Sun, Yuying Chen et al.

In autonomous driving, navigation through unsignaled intersections with many traffic participants moving around is a challenging task. To provide a solution to this problem, we propose a novel branched network G-CIL for the navigation policy learning. Specifically, we firstly represent such dynamic environments as graph-structured data and propose an effective strategy for edge definition to aggregate surrounding information for the ego-vehicle. Then graph convolutional neural networks are used as the perception module to capture global and geometric features from the environment. To generate safe and efficient navigation policy, we further incorporate it with conditional imitation learning algorithm, to learn driving behaviors directly from expert demonstrations. Our proposed network is capable of handling a varying number of surrounding vehicles and generating optimal control actions (e.g., steering angle and throttle) according to the given high-level commands (e.g., turn left towards the global goal). Evaluations on unsignaled intersections with various traffic densities demonstrate that our end-to-end trainable neural network outperforms the baselines with higher success rate and shorter navigation time.

Peide Cai, Hengli Wang, Yuxiang Sun et al.

Traditional decision and planning frameworks for self-driving vehicles (SDVs) scale poorly in new scenarios, thus they require tedious hand-tuning of rules and parameters to maintain acceptable performance in all foreseeable cases. Recently, self-driving methods based on deep learning have shown promising results with better generalization capability but less hand engineering effort. However, most of the previous learning-based methods are trained and evaluated in limited driving scenarios with scattered tasks, such as lane-following, autonomous braking, and conditional driving. In this paper, we propose a graph-based deep network to achieve scalable self-driving that can handle massive traffic scenarios. Specifically, more than 7,000 km of evaluation is conducted in a high-fidelity driving simulator, in which our method can obey the traffic rules and safely navigate the vehicle in a large variety of urban, rural, and highway environments, including unprotected left turns, narrow roads, roundabouts, and pedestrian-rich intersections. Demonstration videos are available at https://caipeide.github.io/dignet/.

Hengli Wang, Rui Fan, Yuxiang Sun et al.

The joint detection of drivable areas and road anomalies is a crucial task for ground mobile robots. In recent years, many impressive semantic segmentation networks, which can be used for pixel-level drivable area and road anomaly detection, have been developed. However, the detection accuracy still needs improvement. Therefore, we develop a novel module named the Normal Inference Module (NIM), which can generate surface normal information from dense depth images with high accuracy and efficiency. Our NIM can be deployed in existing convolutional neural networks (CNNs) to refine the segmentation performance. To evaluate the effectiveness and robustness of our NIM, we embed it in twelve state-of-the-art CNNs. The experimental results illustrate that our NIM can greatly improve the performance of the CNNs for drivable area and road anomaly detection. Furthermore, our proposed NIM-RTFNet ranks 8th on the KITTI road benchmark and exhibits a real-time inference speed.

Hengli Wang, Yuxiang Sun, Ming Liu

The segmentation of drivable areas and road anomalies are critical capabilities to achieve autonomous navigation for robotic wheelchairs. The recent progress of semantic segmentation using deep learning techniques has presented effective results. However, the acquisition of large-scale datasets with hand-labeled ground truth is time-consuming and labor-intensive, making the deep learning-based methods often hard to implement in practice. We contribute to the solution of this problem for the task of drivable area and road anomaly segmentation by proposing a self-supervised learning approach. We develop a pipeline that can automatically generate segmentation labels for drivable areas and road anomalies. Then, we train RGB-D data-based semantic segmentation neural networks and get predicted labels. Experimental results show that our proposed automatic labeling pipeline achieves an impressive speed-up compared to manual labeling. In addition, our proposed self-supervised approach exhibits more robust and accurate results than the state-of-the-art traditional algorithms as well as the state-of-the-art self-supervised algorithms.

Huaiyang Huang, Haoyang Ye, Yuxiang Sun et al.

Incorporating prior structure information into the visual state estimation could generally improve the localization performance. In this letter, we aim to address the paradox between accuracy and efficiency in coupling visual factors with structure constraints. To this end, we present a cross-modality method that tracks a camera in a prior map modelled by the Gaussian Mixture Model (GMM). With the pose estimated by the front-end initially, the local visual observations and map components are associated efficiently, and the visual structure from the triangulation is refined simultaneously. By introducing the hybrid structure factors into the joint optimization, the camera poses are bundle-adjusted with the local visual structure. By evaluating our complete system, namely GMMLoc, on the public dataset, we show how our system can provide a centimeter-level localization accuracy with only trivial computational overhead. In addition, the comparative studies with the state-of-the-art vision-dominant state estimators demonstrate the competitive performance of our method.

Peide Cai, Sukai Wang, Yuxiang Sun et al.

All-day and all-weather navigation is a critical capability for autonomous driving, which requires proper reaction to varied environmental conditions and complex agent behaviors. Recently, with the rise of deep learning, end-to-end control for autonomous vehicles has been well studied. However, most works are solely based on visual information, which can be degraded by challenging illumination conditions such as dim light or total darkness. In addition, they usually generate and apply deterministic control commands without considering the uncertainties in the future. In this paper, based on imitation learning, we propose a probabilistic driving model with ultiperception capability utilizing the information from the camera, lidar and radar. We further evaluate its driving performance online on our new driving benchmark, which includes various environmental conditions (e.g., urban and rural areas, traffic densities, weather and times of the day) and dynamic obstacles (e.g., vehicles, pedestrians, motorcyclists and bicyclists). The results suggest that our proposed model outperforms baselines and achieves excellent generalization performance in unseen environments with heavy traffic and extreme weather.

Peide Cai, Yuxiang Sun, Hengli Wang et al.

Traditional methods for autonomous driving are implemented with many building blocks from perception, planning and control, making them difficult to generalize to varied scenarios due to complex assumptions and interdependencies. Recently, the end-to-end driving method has emerged, which performs well and generalizes to new environments by directly learning from export-provided data. However, many existing methods on this topic neglect to check the confidence of the driving actions and the ability to recover from driving mistakes. In this paper, we develop an uncertainty-aware end-to-end trajectory generation method based on imitation learning. It can extract spatiotemporal features from the front-view camera images for scene understanding, and then generate collision-free trajectories several seconds into the future. The experimental results suggest that under various weather and lighting conditions, our network can reliably generate trajectories in different urban environments, such as turning at intersections and slowing down for collision avoidance. Furthermore, closed-loop driving tests suggest that the proposed method achieves better cross-scene/platform driving results than the state-of-the-art (SOTA) end-to-end control method, where our model can recover from off-center and off-orientation errors and capture 80% of dangerous cases with high uncertainty estimations.

Tianyu Liu, Qinghai Liao, Lu Gan et al.

Since early 2020, the coronavirus disease 2019 (COVID-19) has spread rapidly across the world. As at the date of writing this article, the disease has been globally reported in 223 countries and regions, infected over 108 million people and caused over 2.4 million deaths (https://covid19.who.int/, accessed on Feb. 17, 2021). Avoiding person-to-person transmission is an effective approach to control and prevent the pandemic. However, many daily activities, such as transporting goods in our daily life, inevitably involve person-to-person contact. Using an autonomous logistic vehicle to achieve contact-less goods transportation could alleviate this issue. For example, it can reduce the risk of virus transmission between the driver and customers. Moreover, many countries have imposed tough lockdown measures to reduce the virus transmission (e.g., retail, catering) during the pandemic, which causes inconveniences for human daily life. Autonomous vehicle can deliver the goods bought by humans, so that humans can get the goods without going out. These demands motivate us to develop an autonomous vehicle, named as Hercules, for contact-less goods transportation during the COVID-19 pandemic. The vehicle is evaluated through real-world delivering tasks under various traffic conditions.

Huaiyang Huang, Yuxiang Sun, Haoyang Ye et al.

Metric localization plays a critical role in vision-based navigation. For overcoming the degradation of matching photometry under appearance changes, recent research resorted to introducing geometry constraints of the prior scene structure. In this paper, we present a metric localization method for the monocular camera, using the Signed Distance Field (SDF) as a global map representation. Leveraging the volumetric distance information from SDFs, we aim to relax the assumption of an accurate structure from the local Bundle Adjustment (BA) in previous methods. By tightly coupling the distance factor with temporal visual constraints, our system corrects the odometry drift and jointly optimizes global camera poses with the local structure. We validate the proposed approach on both indoor and outdoor public datasets. Compared to the state-of-the-art methods, it achieves a comparable performance with a minimal sensor configuration.

Sukai Wang, Yuxiang Sun, Chengju Liu et al.

Recent machine learning-based multi-object tracking (MOT) frameworks are becoming popular for 3-D point clouds. Most traditional tracking approaches use filters (e.g., Kalman filter or particle filter) to predict object locations in a time sequence, however, they are vulnerable to extreme motion conditions, such as sudden braking and turning. In this letter, we propose PointTrackNet, an end-to-end 3-D object detection and tracking network, to generate foreground masks, 3-D bounding boxes, and point-wise tracking association displacements for each detected object. The network merely takes as input two adjacent point-cloud frames. Experimental results on the KITTI tracking dataset show competitive results over the state-of-the-arts, especially in the irregularly and rapidly changing scenarios.

Peide Cai, Xiaodong Mei, Lei Tai et al.

Drifting is a complicated task for autonomous vehicle control. Most traditional methods in this area are based on motion equations derived by the understanding of vehicle dynamics, which is difficult to be modeled precisely. We propose a robust drift controller without explicit motion equations, which is based on the latest model-free deep reinforcement learning algorithm soft actor-critic. The drift control problem is formulated as a trajectory following task, where the errorbased state and reward are designed. After being trained on tracks with different levels of difficulty, our controller is capable of making the vehicle drift through various sharp corners quickly and stably in the unseen map. The proposed controller is further shown to have excellent generalization ability, which can directly handle unseen vehicle types with different physical properties, such as mass, tire friction, etc.

Yuxiang Sun, Pooyan Fazli

Policy distillation in deep reinforcement learning provides an effective way to transfer control policies from a larger network to a smaller untrained network without a significant degradation in performance. However, policy distillation is underexplored in deep reinforcement learning, and existing approaches are computationally inefficient, resulting in a long distillation time. In addition, the effectiveness of the distillation process is still limited to the model capacity. We propose a new distillation mechanism, called real-time policy distillation, in which training the teacher model and distilling the policy to the student model occur simultaneously. Accordingly, the teacher's latest policy is transferred to the student model in real time. This reduces the distillation time to half the original time or even less and also makes it possible for extremely small student models to learn skills at the expert level. We evaluated the proposed algorithm in the Atari 2600 domain. The results show that our approach can achieve full distillation in most games, even with compression ratios up to 1.7%.

Umar Ozgunalp, Rui Fan, Shanshan Cheng et al.

In this paper, a robust lane detection algorithm is proposed, where the vertical road profile of the road is estimated using dynamic programming from the v-disparity map and, based on the estimated profile, the road area is segmented. Since the lane markings are on the road area and any feature point above the ground will be a noise source for the lane detection, a mask is created for the road area to remove some of the noise for lane detection. The estimated mask is multiplied by the lane feature map in a bird's eye view (BEV). The lane feature points are extracted by using an extended version of symmetrical local threshold (SLT), which not only considers dark light dark transition (DLD) of the lane markings, like (SLT), but also considers parallelism on the lane marking borders. The segmentation then uses only the feature points that are on the road area. A maximum of two linear lane markings are detected using an efficient 1D Hough transform. Then, the detected linear lane markings are used to create a region of interest (ROI) for parabolic lane detection. Finally, based on the estimated region of interest, parabolic lane models are fitted using robust fitting. Due to the robust lane feature extraction and road area segmentation, the proposed algorithm robustly detects lane markings and achieves lane marking detection with an accuracy of 91% when tested on a sequence from the KITTI dataset.

Ting Sun, Yuxiang Sun, Ming Liu et al.

Moving objects can greatly jeopardize the performance of a visual simultaneous localization and mapping (vSLAM) system which relies on the static-world assumption. Motion removal have seen successful on solving this problem. Two main streams of solutions are based on either geometry constraints or deep semantic segmentation neural network. The former rely on static majority assumption, and the latter require labor-intensive pixel-wise annotations. In this paper we propose to adopt a novel weakly-supervised semantic segmentation method. The segmentation mask is obtained from a CNN pre-trained with image-level class labels only. Thus, we leverage the power of deep semantic segmentation CNNs, while avoid requiring expensive annotations for training. We integrate our motion removal approach with the ORB-SLAM2 system. Experimental results on the TUM RGB-D and the KITTI stereo datasets demonstrate our superiority over the state-of-the-art.