Junqiao Zhao

Gengxuan Tian, Junqiao Zhao, Yingfeng Cai et al.

LiDAR-based place recognition plays a crucial role in Simultaneous Localization and Mapping (SLAM) and LiDAR localization. Despite the emergence of various deep learning-based and hand-crafting-based methods, rotation-induced place recognition failure remains a critical challenge. Existing studies address this limitation through specific training strategies or network structures. However, the former does not produce satisfactory results, while the latter focuses mainly on the reduced problem of SO(2) rotation invariance. Methods targeting SO(3) rotation invariance suffer from limitations in discrimination capability. In this paper, we propose a new method that employs Vector Neurons Network (VNN) to achieve SO(3) rotation invariance. We first extract rotation-equivariant features from neighboring points and map low-dimensional features to a high-dimensional space through VNN. Afterwards, we calculate the Euclidean and Cosine distance in the rotation-equivariant feature space as rotation-invariant feature descriptors. Finally, we aggregate the features using GeM pooling to obtain global descriptors. To address the significant information loss when formulating rotation-invariant descriptors, we propose computing distances between features at different layers within the Euclidean space neighborhood. This greatly improves the discriminability of the point cloud descriptors while ensuring computational efficiency. Experimental results on public datasets show that our approach significantly outperforms other baseline methods implementing rotation invariance, while achieving comparable results with current state-of-the-art place recognition methods that do not consider rotation issues.

Junqi Liu, Zihao Zhou, Zekai Zhu et al.

Agentic systems have recently become the dominant paradigm for formal theorem proving, achieving strong performance by coordinating multiple models and tools. However, existing approaches often rely on task-specific pipelines and trained formal provers, limiting their flexibility and reproducibility. In this paper, we propose the paradigm that directly uses a general coding agent as a formal math reasoner. This paradigm is motivated by (1) A general coding agent provides a natural interface for diverse reasoning tasks beyond proving, (2) Performance can be improved by simply replacing the underlying base model, without training, and (3) MCP enables flexible extension and autonomous calling of specialized tools, avoiding complex design. Based on this paradigm, we introduce Numina-Lean-Agent, which combines Claude Code with Numina-Lean-MCP to enable autonomous interaction with Lean, retrieval of relevant theorems, informal proving and auxiliary reasoning tools. Using Claude Opus 4.5 as the base model, Numina-Lean-Agent solves all problems in Putnam 2025 (12 / 12), matching the best closed-source system. Beyond benchmark evaluation, we further demonstrate its generality by interacting with mathematicians to successfully formalize the Brascamp-Lieb theorem. We release Numina-Lean-Agent and all solutions at https://github.com/project-numina/numina-lean-agent.

Hai Zhang, Hang Yu, Junqiao Zhao et al.

Designing and deriving effective model-based reinforcement learning (MBRL) algorithms with a performance improvement guarantee is challenging, mainly attributed to the high coupling between model learning and policy optimization. Many prior methods that rely on return discrepancy to guide model learning ignore the impacts of model shift, which can lead to performance deterioration due to excessive model updates. Other methods use performance difference bound to explicitly consider model shift. However, these methods rely on a fixed threshold to constrain model shift, resulting in a heavy dependence on the threshold and a lack of adaptability during the training process. In this paper, we theoretically derive an optimization objective that can unify model shift and model bias and then formulate a fine-tuning process. This process adaptively adjusts the model updates to get a performance improvement guarantee while avoiding model overfitting. Based on these, we develop a straightforward algorithm USB-PO (Unified model Shift and model Bias Policy Optimization). Empirical results show that USB-PO achieves state-of-the-art performance on several challenging benchmark tasks.

Xiao Zhang, Hai Zhang, Hongtu Zhou et al.

Safety is one of the main challenges in applying reinforcement learning to realistic environmental tasks. To ensure safety during and after training process, existing methods tend to adopt overly conservative policy to avoid unsafe situations. However, overly conservative policy severely hinders the exploration, and makes the algorithms substantially less rewarding. In this paper, we propose a method to construct a boundary that discriminates safe and unsafe states. The boundary we construct is equivalent to distinguishing dead-end states, indicating the maximum extent to which safe exploration is guaranteed, and thus has minimum limitation on exploration. Similar to Recovery Reinforcement Learning, we utilize a decoupled RL framework to learn two policies, (1) a task policy that only considers improving the task performance, and (2) a recovery policy that maximizes safety. The recovery policy and a corresponding safety critic are pretrained on an offline dataset, in which the safety critic evaluates upper bound of safety in each state as awareness of environmental safety for the agent. During online training, a behavior correction mechanism is adopted, ensuring the agent to interact with the environment using safe actions only. Finally, experiments of continuous control tasks demonstrate that our approach has better task performance with less safety violations than state-of-the-art algorithms.

Wenjie Mu, Zhan Li, Chuanzhou Su et al.

Generalizable Neural Radiance Fields (GeNeRFs) enable high-quality scene reconstruction from sparse views and can generalize to unseen scenes. However, in real-world settings, transient distractors break cross-view structural consistency, corrupting supervision and degrading reconstruction quality. Existing distractor-free NeRF methods rely on per-scene optimization and estimate uncertainty from per-view reconstruction errors, which are not reliable for GeNeRFs and often misjudge inconsistent static structures as distractors. To this end, we propose MU-GeNeRF, a Multi-view Uncertainty-guided distractor-aware GeNeRF framework designed to alleviate GeNeRF's robust modeling challenges in the presence of transient distractions. We decompose distractor awareness into two complementary uncertainty components: Source-view Uncertainty, which captures structural discrepancies across source views caused by viewpoint changes or dynamic factors; and Target-view Uncertainty, which detects observation anomalies in the target image induced by transient distractors.These two uncertainties address distinct error sources and are combined through a heteroscedastic reconstruction loss, which guides the model to adaptively modulate supervision, enabling more robust distractor suppression and geometric modeling.Extensive experiments show that our method not only surpasses existing GeNeRFs but also achieves performance comparable to scene-specific distractor-free NeRFs.

Hang Yu, Juntu Zhao, Yufeng Liu et al.

Vision-Language-Action (VLA) models align vision and language with embodied control, but their object referring ability remains limited when relying solely on text prompt, especially in cluttered or out-of-distribution (OOD) scenes. In this study, we introduce the Point-VLA, a plug-and-play policy that augments language instructions with explicit visual cues (e.g., bounding boxes) to resolve referential ambiguity and enable precise object-level grounding. To efficiently scale visually grounded datasets, we further develop an automatic data annotation pipeline requiring minimal human effort. We evaluate Point-VLA on diverse real-world referring tasks and observe consistently stronger performance than text-only instruction VLAs, particularly in cluttered or unseen-object scenarios, with robust generalization. These results demonstrate that Point-VLA effectively resolves object referring ambiguity through pixel-level visual grounding, achieving more generalizable embodied control.

Di Zhang, Bowen Lv, Hai Zhang et al.

A primary challenge for visual-based Reinforcement Learning (RL) is to generalize effectively across unseen environments. Although previous studies have explored different auxiliary tasks to enhance generalization, few adopt image reconstruction due to concerns about exacerbating overfitting to task-irrelevant features during training. Perceiving the pre-eminence of image reconstruction in representation learning, we propose SMG (Separated Models for Generalization), a novel approach that exploits image reconstruction for generalization. SMG introduces two model branches to extract task-relevant and task-irrelevant representations separately from visual observations via cooperatively reconstruction. Built upon this architecture, we further emphasize the importance of task-relevant features for generalization. Specifically, SMG incorporates two additional consistency losses to guide the agent's focus toward task-relevant areas across different scenarios, thereby achieving free from overfitting. Extensive experiments in DMC demonstrate the SOTA performance of SMG in generalization, particularly excelling in video-background settings. Evaluations on robotic manipulation tasks further confirm the robustness of SMG in real-world applications.

Xufei Wang, Junqiao Zhao, Siyue Tao et al.

LiDAR place recognition plays a crucial role in SLAM, robot navigation, and autonomous driving. However, existing LiDAR place recognition methods often struggle to adapt to new environments without forgetting previously learned knowledge, a challenge widely known as catastrophic forgetting. To address this issue, we propose KDF+, a novel continual learning framework for LiDAR place recognition that extends the KDF paradigm with a loss-aware sampling strategy and a rehearsal enhancement mechanism. The proposed sampling strategy estimates the learning difficulty of each sample via its loss value and selects samples for replay according to their estimated difficulty. Harder samples, which tend to encode more discriminative information, are sampled with higher probability while maintaining distributional coverage across the dataset. In addition, the rehearsal enhancement mechanism encourages memory samples to be further refined during new-task training by slightly reducing their loss relative to previous tasks, thereby reinforcing long-term knowledge retention. Extensive experiments across multiple benchmarks demonstrate that KDF+ consistently outperforms existing continual learning methods and can be seamlessly integrated into state-of-the-art continual learning for LiDAR place recognition frameworks to yield significant and stable performance gains. The code will be available at https://github.com/repo/KDF-plus.

Qian Chen, Junqiao Zhao, Hongtu Zhou et al.

Long-horizon, sparse-reward tasks pose a fundamental challenge for reinforcement learning, since single-step TD learning suffers from bootstrapping error accumulation across successive Bellman updates. Actor-critic methods with action chunking address this by operating over temporally extended actions, which reduce the effective horizon, enable fast value backups, and support temporally consistent exploration. However, existing methods rely on a fixed chunk size and therefore cannot adaptively balance reactivity against temporal consistency. A large fixed chunk size reduces responsiveness to new observations, while a small one produces incoherent motions, forcing task-specific tuning of the chunk size. To address this limitation, we propose Adaptive Chunk Size Actor-Critic (ACSAC). ACSAC leverages a causal Transformer critic to evaluate expected returns for action chunks of different sizes. At each chunk boundary, it adaptively selects the chunk size that maximizes the expected return, supporting flexible, state-dependent chunk sizes without task-specific tuning. We prove that the ACSAC Bellman operator is a contraction whose unique fixed point is the action-value function of the adaptive policy. Experiments on OGBench demonstrate that ACSAC achieves state-of-the-art performance on long-horizon, sparse-reward manipulation tasks across both offline RL and offline-to-online RL settings.

Xiaolong Yu, Junqiao Zhao, Shuangfu Song et al.

Using Quadrics as the object representation has the benefits of both generality and closed-form projection derivation between image and world spaces. Although numerous constraints have been proposed for dual quadric reconstruction, we found that many of them are imprecise and provide minimal improvements to localization.After scrutinizing the existing constraints, we introduce a concise yet more precise convex hull-based algebraic constraint for object landmarks, which is applied to object reconstruction, frontend pose estimation, and backend bundle adjustment.This constraint is designed to fully leverage precise semantic segmentation, effectively mitigating mismatches between complex-shaped object contours and dual quadrics.Experiments on public datasets demonstrate that our approach is applicable to both monocular and RGB-D SLAM and achieves improved object mapping and localization than existing quadric SLAM methods. The implementation of our method is available at https://github.com/tiev-tongji/convexhull-based-algebraic-constraint.

Junqiao Zhao, Fenglin Zhang, Yingfeng Cai et al.

Visual Place Recognition (VPR) aims to retrieve frames from a geotagged database that are located at the same place as the query frame. To improve the robustness of VPR in perceptually aliasing scenarios, sequence-based VPR methods are proposed. These methods are either based on matching between frame sequences or extracting sequence descriptors for direct retrieval. However, the former is usually based on the assumption of constant velocity, which is difficult to hold in practice, and is computationally expensive and subject to sequence length. Although the latter overcomes these problems, existing sequence descriptors are constructed by aggregating features of multiple frames only, without interaction on temporal information, and thus cannot obtain descriptors with spatio-temporal discrimination.In this paper, we propose a sequence descriptor that effectively incorporates spatio-temporal information. Specifically, spatial attention within the same frame is utilized to learn spatial feature patterns, while attention in corresponding local regions of different frames is utilized to learn the persistence or change of features over time. We use a sliding window to control the temporal range of attention and use relative positional encoding to construct sequential relationships between different features. This allows our descriptors to capture the intrinsic dynamics in a sequence of frames.Comprehensive experiments on challenging benchmark datasets show that the proposed approach outperforms recent state-of-the-art methods.The code is available at https://github.com/tiev-tongji/Spatio-Temporal-SeqVPR.

Jianfeng Li, Junqiao Zhao, Shuangfu Song et al.

Estimating geometric elements such as depth, camera motion, and optical flow from images is an important part of the robot's visual perception. We use a joint self-supervised method to estimate the three geometric elements. Depth network, optical flow network and camera motion network are independent of each other but are jointly optimized during training phase. Compared with independent training, joint training can make full use of the geometric relationship between geometric elements and provide dynamic and static information of the scene. In this paper, we improve the joint self-supervision method from three aspects: network structure, dynamic object segmentation, and geometric constraints. In terms of network structure, we apply the attention mechanism to the camera motion network, which helps to take advantage of the similarity of camera movement between frames. And according to attention mechanism in Transformer, we propose a plug-and-play convolutional attention module. In terms of dynamic object, according to the different influences of dynamic objects in the optical flow self-supervised framework and the depth-pose self-supervised framework, we propose a threshold algorithm to detect dynamic regions, and mask that in the loss function respectively. In terms of geometric constraints, we use traditional methods to estimate the fundamental matrix from the corresponding points to constrain the camera motion network. We demonstrate the effectiveness of our method on the KITTI dataset. Compared with other joint self-supervised methods, our method achieves state-of-the-art performance in the estimation of pose and optical flow, and the depth estimation has also achieved competitive results. Code will be available https://github.com/jianfenglihg/Unsupervised_geometry.

Jianfeng Li, Junqiao Zhao, Tiantian Feng et al.

In this paper, we proposed an unsupervised learning method for estimating the optical flow between video frames, especially to solve the occlusion problem. Occlusion is caused by the movement of an object or the movement of the camera, defined as when certain pixels are visible in one video frame but not in adjacent frames. Due to the lack of pixel correspondence between frames in the occluded area, incorrect photometric loss calculation can mislead the optical flow training process. In the video sequence, we found that the occlusion in the forward ($t\rightarrow t+1$) and backward ($t\rightarrow t-1$) frame pairs are usually complementary. That is, pixels that are occluded in subsequent frames are often not occluded in the previous frame and vice versa. Therefore, by using this complementarity, a new weighted loss is proposed to solve the occlusion problem. In addition, we calculate gradients in multiple directions to provide richer supervision information. Our method achieves competitive optical flow accuracy compared to the baseline and some supervised methods on KITTI 2012 and 2015 benchmarks. This source code has been released at https://github.com/jianfenglihg/UnOpticalFlow.git.

Qingjun Wang, Hongtu Zhou, Hang Yu et al.

Offline reinforcement learning (RL) faces a critical challenge of overestimating the value of out-of-distribution (OOD) actions. Existing methods mitigate this issue by penalizing unseen samples, yet they fail to accurately identify OOD actions and may suppress beneficial exploration beyond the behavioral support. Although several methods have been proposed to differentiate OOD samples with distinct properties, they typically rely on restrictive assumptions about the data distribution and remain limited in discrimination ability. To address this problem, we propose DOSER (Diffusion-based OOD Detection and Selective Regularization), a novel framework that goes beyond uniform penalization. DOSER trains two diffusion models to capture the behavior policy and state distribution, using single-step denoising reconstruction error as a reliable OOD indicator. During policy optimization, it further distinguishes between beneficial and detrimental OOD actions by evaluating predicted transitions, selectively suppressing risky actions while encouraging exploration of high-potential ones. Theoretically, we prove that DOSER is a $γ$-contraction and therefore admits a unique fixed point with bounded value estimates. We further provide an asymptotic performance guarantee relative to the optimal policy under model approximation and OOD detection errors. Across extensive offline RL benchmarks, DOSER consistently attains superior performance to prior methods, especially on suboptimal datasets.

Lanqing Li, Hai Zhang, Xinyu Zhang et al.

As a marriage between offline RL and meta-RL, the advent of offline meta-reinforcement learning (OMRL) has shown great promise in enabling RL agents to multi-task and quickly adapt while acquiring knowledge safely. Among which, context-based OMRL (COMRL) as a popular paradigm, aims to learn a universal policy conditioned on effective task representations. In this work, by examining several key milestones in the field of COMRL, we propose to integrate these seemingly independent methodologies into a unified framework. Most importantly, we show that the pre-existing COMRL algorithms are essentially optimizing the same mutual information objective between the task variable $M$ and its latent representation $Z$ by implementing various approximate bounds. Such theoretical insight offers ample design freedom for novel algorithms. As demonstrations, we propose a supervised and a self-supervised implementation of $I(Z; M)$, and empirically show that the corresponding optimization algorithms exhibit remarkable generalization across a broad spectrum of RL benchmarks, context shift scenarios, data qualities and deep learning architectures. This work lays the information theoretic foundation for COMRL methods, leading to a better understanding of task representation learning in the context of reinforcement learning. Given its generality, we envision our framework as a promising offline pre-training paradigm of foundation models for decision making.

Xufei Wang, Gengxuan Tian, Junqiao Zhao et al.

Place recognition plays a significant role in SLAM, robot navigation, and autonomous driving applications. Benefiting from deep learning, the performance of LiDAR place recognition (LPR) has been greatly improved. However, many existing learning-based LPR methods suffer from catastrophic forgetting, which severely harms the performance of LPR on previously trained places after training on a new environment. In this paper, we introduce a continual learning framework for LPR via Knowledge Distillation and Fusion (KDF) to alleviate forgetting. Inspired by the ranking process of place recognition retrieval, we present a ranking-aware knowledge distillation loss that encourages the network to preserve the high-level place recognition knowledge. We also introduce a knowledge fusion module to integrate the knowledge of old and new models for LiDAR place recognition. Our extensive experiments demonstrate that KDF can be applied to different networks to overcome catastrophic forgetting, surpassing the state-of-the-art methods in terms of mean Recall@1 and forgetting score.

Jijun Wang, Yan Wu, Yujian Mo et al.

Existing LiDAR-Camera fusion methods have achieved strong results in 3D object detection. To address the sparsity of point clouds, previous approaches typically construct spatial pseudo point clouds via depth completion as auxiliary input and adopts a proposal-refinement framework to generate detection results. However, introducing pseudo points inevitably brings noise, potentially resulting in inaccurate predictions. Considering the differing roles and reliability levels of each modality, we propose LDRFusion, a novel Lidar-dominant two-stage refinement framework for multi-sensor fusion. The first stage soley relies on LiDAR to produce accurately localized proposals, followed by a second stage where pseudo point clouds are incorporated to detect challenging instances. The instance-level results from both stages are subsequently merged. To further enhance the representation of local structures in pseudo point clouds, we present a hierarchical pseudo point residual encoding module, which encodes neighborhood sets using both feature and positional residuals. Experiments on the KITTI dataset demonstrate that our framework consistently achieves strong performance across multiple categories and difficulty levels.

Yujian Mo, Yan Wu, Junqiao Zhao et al.

Recent advances in foundation models have opened up new possibilities for enhancing 3D perception. In particular, DepthAnything offers dense and reliable geometric priors from monocular RGB images, which can complement sparse LiDAR data in autonomous driving scenarios. However, such priors remain underutilized in LiDAR-based 3D object detection. In this paper, we address the limited expressiveness of raw LiDAR point features, especially the weak discriminative capability of the reflectance attribute, by introducing depth priors predicted by DepthAnything. These priors are fused with the original LiDAR attributes to enrich each point's representation. To leverage the enhanced point features, we propose a point-wise feature extraction module. Then, a Dual-Path RoI feature extraction framework is employed, comprising a voxel-based branch for global semantic context and a point-based branch for fine-grained structural details. To effectively integrate the complementary RoI features, we introduce a bidirectional gated RoI feature fusion module that balances global and local cues. Extensive experiments on the KITTI benchmark show that our method consistently improves detection accuracy, demonstrating the value of incorporating visual foundation model priors into LiDAR-based 3D object detection.

Hang Yu, Di Zhang, Qiwei Du et al.

Offline reinforcement learning (RL) enables agents to learn optimal policies from pre-collected datasets. However, datasets containing suboptimal and fragmented trajectories present challenges for reward propagation, resulting in inaccurate value estimation and degraded policy performance. While trajectory stitching via generative models offers a promising solution, existing augmentation methods frequently produce trajectories that are either confined to the support of the behavior policy or violate the underlying dynamics, thereby limiting their effectiveness for policy improvement. We propose ASTRO, a data augmentation framework that generates distributionally novel and dynamics-consistent trajectories for offline RL. ASTRO first learns a temporal-distance representation to identify distinct and reachable stitch targets. We then employ a dynamics-guided stitch planner that adaptively generates connecting action sequences via Rollout Deviation Feedback, defined as the gap between target state sequence and the actual arrived state sequence by executing predicted actions, to improve trajectory stitching's feasibility and reachability. This approach facilitates effective augmentation through stitching and ultimately enhances policy learning. ASTRO outperforms prior offline RL augmentation methods across various algorithms, achieving notable performance gain on the challenging OGBench suite and demonstrating consistent improvements on standard offline RL benchmarks such as D4RL.

Xinyu Zhang, Kai Huang, Junqiao Zhao et al.

We propose a multi-camera LiDAR-visual-inertial odometry framework, Multi-LVI-SAM, which fuses data from multiple fisheye cameras, LiDAR and inertial sensors for highly accurate and robust state estimation. To enable efficient and consistent integration of visual information from multiple fisheye cameras, we introduce a panoramic visual feature model that unifies multi-camera observations into a single representation. The panoramic model serves as a global geometric optimization framework that consolidates multi-view constraints, enabling seamless loop closure and global pose optimization, while simplifying system design by avoiding redundant handling of individual cameras. To address the triangulation inconsistency caused by the misalignment between each camera's frame and the panoramic model's frame, we propose an extrinsic compensation method. This method improves feature consistency across views and significantly reduces triangulation and optimization errors, leading to more accurate pose estimation. We integrate the panoramic visual feature model into a tightly coupled LiDAR-visual-inertial system based on a factor graph. Extensive experiments on public datasets demonstrate that the panoramic visual feature model enhances the quality and consistency of multi-camera constraints, resulting in higher accuracy and robustness than existing multi-camera LiDAR-visual-inertial systems.

Qiwen Gu, Xufei Wang, Junqiao Zhao et al.

Visual Place Recognition (VPR) enables robust localization through image retrieval based on learned descriptors. However, drastic appearance variations of images at the same place caused by viewpoint changes can lead to inconsistent supervision signals, thereby degrading descriptor learning. Existing methods either rely on manually defined cropping rules or labeled data for view differentiation, but they suffer from two major limitations: (1) reliance on labels or handcrafted rules restricts generalization capability; (2) even within the same view direction, occlusions can introduce feature ambiguity. To address these issues, we propose MutualVPR, a mutual learning framework that integrates unsupervised view self-classification and descriptor learning. We first group images by geographic coordinates, then iteratively refine the clusters using K-means to dynamically assign place categories without orientation labels. Specifically, we adopt a DINOv2-based encoder to initialize the clustering. During training, the encoder and clustering co-evolve, progressively separating drastic appearance variations of the same place and enabling consistent supervision. Furthermore, we find that capturing fine-grained image differences at a place enhances robustness. Experiments demonstrate that MutualVPR achieves state-of-the-art (SOTA) performance across multiple datasets, validating the effectiveness of our framework in improving view direction generalization, occlusion robustness.

Hai Zhang, Boyuan Zheng, Tianying Ji et al.

Offline meta reinforcement learning (OMRL) has emerged as a promising approach for interaction avoidance and strong generalization performance by leveraging pre-collected data and meta-learning techniques. Previous context-based approaches predominantly rely on the intuition that alternating optimization between the context encoder and the policy can lead to performance improvements, as long as the context encoder follows the principle of maximizing the mutual information between the task variable $M$ and its latent representation $Z$ ($I(Z;M)$) while the policy adopts the standard offline reinforcement learning (RL) algorithms conditioning on the learned task representation.Despite promising results, the theoretical justification of performance improvements for such intuition remains underexplored.Inspired by the return discrepancy scheme in the model-based RL field, we find that the previous optimization framework can be linked with the general RL objective of maximizing the expected return, thereby explaining performance improvements. Furthermore, after scrutinizing this optimization framework, we observe that the condition for monotonic performance improvements does not consider the variation of the task representation. When these variations are considered, the previously established condition may no longer be sufficient to ensure monotonicity, thereby impairing the optimization process.We name this issue task representation shift and theoretically prove that the monotonic performance improvements can be guaranteed with appropriate context encoder updates.Our work opens up a new avenue for OMRL, leading to a better understanding between the task representation and performance improvements.

Chang Huang, Shatong Zhu, Junqiao Zhao et al.

Value function factorization methods are commonly used in cooperative multi-agent reinforcement learning, with QMIX receiving significant attention. Many QMIX-based methods introduce monotonicity constraints between the joint action value and individual action values to achieve decentralized execution. However, such constraints limit the representation capacity of value factorization, restricting the joint action values it can represent and hindering the learning of the optimal policy. To address this challenge, we propose the Potentially Optimal Joint Actions Weighted QMIX (POWQMIX) algorithm, which recognizes the potentially optimal joint actions and assigns higher weights to the corresponding losses of these joint actions during training. We theoretically prove that with such a weighted training approach the optimal policy is guaranteed to be recovered. Experiments in matrix games, difficulty-enhanced predator-prey, and StarCraft II Multi-Agent Challenge environments demonstrate that our algorithm outperforms the state-of-the-art value-based multi-agent reinforcement learning methods.

Xuebo Tian, Junqiao Zhao, Chen Ye

Ego-pose estimation and dynamic object tracking are two key issues in an autonomous driving system. Two assumptions are often made for them, i.e. the static world assumption of simultaneous localization and mapping (SLAM) and the exact ego-pose assumption of object tracking, respectively. However, these assumptions are difficult to hold in highly dynamic road scenarios where SLAM and object tracking become correlated and mutually beneficial. In this paper, DL-SLOT, a dynamic Lidar SLAM and object tracking method is proposed. This method integrates the state estimations of both the ego vehicle and the static and dynamic objects in the environment into a unified optimization framework, to realize SLAM and object tracking (SLOT) simultaneously. Firstly, we implement object detection to remove all the points that belong to potential dynamic objects. Then, LiDAR odometry is conducted using the filtered point cloud. At the same time, detected objects are associated with the history object trajectories based on the time-series information in a sliding window. The states of the static and dynamic objects and ego vehicle in the sliding window are integrated into a unified local optimization framework. We perform SLAM and object tracking simultaneously in this framework, which significantly improves the robustness and accuracy of SLAM in highly dynamic road scenarios and the accuracy of objects' states estimation. Experiments on public datasets have shown that our method achieves better accuracy than A-LOAM.

Yingfeng Cai, Junqiao Zhao, Jiafeng Cui et al.

Visual Place Recognition (VPR) in areas with similar scenes such as urban or indoor scenarios is a major challenge. Existing VPR methods using global descriptors have difficulty capturing local specific regions (LSR) in the scene and are therefore prone to localization confusion in such scenarios. As a result, finding the LSR that are critical for location recognition becomes key. To address this challenge, we introduced Patch-NetVLAD+, which was inspired by patch-based VPR researches. Our method proposed a fine-tuning strategy with triplet loss to make NetVLAD suitable for extracting patch-level descriptors. Moreover, unlike existing methods that treat all patches in an image equally, our method extracts patches of LSR, which present less frequently throughout the dataset, and makes them play an important role in VPR by assigning proper weights to them. Experiments on Pittsburgh30k and Tokyo247 datasets show that our approach achieved up to 6.35\% performance improvement than existing patch-based methods.

Shuangfu Song, Junqiao Zhao, Tiantian Feng et al.

The scale ambiguity problem is inherently unsolvable to monocular SLAM without the metric baseline between moving cameras. In this paper, we present a novel scale estimation approach based on an object-level SLAM system. To obtain the absolute scale of the reconstructed map, we derive a nonlinear optimization method to make the scaled dimensions of objects conforming to the distribution of their sizes in the physical world, without relying on any prior information of gravity direction. We adopt the dual quadric to represent objects for its ability to fit objects compactly and accurately. In the proposed monocular object-level SLAM system, dual quadrics are fastly initialized based on constraints of 2-D detections and fitted oriented bounding box and are further optimized to provide reliable dimensions for scale estimation.

Jiafeng Cui, Tengfei Huang, Yingfeng Cai et al.

LiDAR-based place recognition is an essential and challenging task both in loop closure detection and global relocalization. We propose Deep Scan Context (DSC), a general and discriminative global descriptor that captures the relationship among segments of a point cloud. Unlike previous methods that utilize either semantics or a sequence of adjacent point clouds for better place recognition, we only use raw point clouds to get competitive results. Concretely, we first segment the point cloud egocentrically to acquire centroids and eigenvalues of the segments. Then, we introduce a graph neural network to aggregate these features into an embedding representation. Extensive experiments conducted on the KITTI dataset show that DSC is robust to scene variants and outperforms existing methods.

Xinneng Yang, Yan Wu, Junqiao Zhao et al.

Semantic segmentation has achieved remarkable results with high computational cost and a large number of parameters. However, real-world applications require efficient inference speed on embedded devices. Most previous works address the challenge by reducing depth, width and layer capacity of network, which leads to poor performance. In this paper, we introduce a novel Dense Dual-Path Network (DDPNet) for real-time semantic segmentation under resource constraints. We design a light-weight and powerful backbone with dense connectivity to facilitate feature reuse throughout the whole network and the proposed Dual-Path module (DPM) to sufficiently aggregate multi-scale contexts. Meanwhile, a simple and effective framework is built with a skip architecture utilizing the high-resolution feature maps to refine the segmentation output and an upsampling module leveraging context information from the feature maps to refine the heatmaps. The proposed DDPNet shows an obvious advantage in balancing accuracy and speed. Specifically, on Cityscapes test dataset, DDPNet achieves 75.3% mIoU with 52.6 FPS for an input of 1024 X 2048 resolution on a single GTX 1080Ti card. Compared with other state-of-the-art methods, DDPNet achieves a significant better accuracy with a comparable speed and fewer parameters.

Zhuoping Yu, Xiaozhou Ren, Yuyao Huang et al.

Accurate detection of lane and road markings is a task of great importance for intelligent vehicles. In existing approaches, the detection accuracy often degrades with the increasing distance. This is due to the fact that distant lane and road markings occupy a small number of pixels in the image, and scales of lane and road markings are inconsistent at various distances and perspectives. The Inverse Perspective Mapping (IPM) can be used to eliminate the perspective distortion, but the inherent interpolation can lead to artifacts especially around distant lane and road markings and thus has a negative impact on the accuracy of lane marking detection and segmentation. To solve this problem, we adopt the Encoder-Decoder architecture in Fully Convolutional Networks and leverage the idea of Spatial Transformer Networks to introduce a novel semantic segmentation neural network. This approach decomposes the IPM process into multiple consecutive differentiable homographic transform layers, which are called "Perspective Transformer Layers". Furthermore, the interpolated feature map is refined by subsequent convolutional layers thus reducing the artifacts and improving the accuracy. The effectiveness of the proposed method in lane marking detection is validated on two public datasets: TuSimple and ApolloScape

Hongtu Zhou, Xinneng Yang, Enwei Zhang et al.

Real-time multi-target path planning is a key issue in the field of autonomous driving. Although multiple paths can be generated in real-time with polynomial curves, the generated paths are not flexible enough to deal with complex road scenes such as S-shaped road and unstructured scenes such as parking lots. Search and sampling-based methods, such as A* and RRT and their derived methods, are flexible in generating paths for these complex road environments. However, the existing algorithms require significant time to plan to multiple targets, which greatly limits their application in autonomous driving. In this paper, a real-time path planning method for multi-targets is proposed. We train a fully convolutional neural network (FCN) to predict a path region for the target at first. By taking the predicted path region as soft constraints, the A* algorithm is then applied to search the exact path to the target. Experiments show that FCN can make multiple predictions in a very short time (50 times in 40ms), and the predicted path region effectively restrict the searching space for the following A* search. Therefore, the A* can search much faster so that the multi-target path planning can be achieved in real-time (3 targets in less than 100ms).

Yewei Huang, Junqiao Zhao, Xudong He et al.

In this paper, we proposed a novel and practical solution for the real-time indoor localization of autonomous driving in parking lots. High-level landmarks, the parking slots, are extracted and enriched with labels to avoid the aliasing of low-level visual features. We then proposed a robust method for detecting incorrect data associations between parking slots and further extended the optimization framework by dynamically eliminating suboptimal data associations. Visual fiducial markers are introduced to improve the overall precision. As a result, a semantic map of the parking lot can be established fully automatically and robustly. We experimented the performance of real-time localization based on the map using our autonomous driving platform TiEV, and the average accuracy of 0.3m track tracing can be achieved at a speed of 10kph.

Lu Sun, Junqiao Zhao, Xudong He et al.

For autonomous vehicles, high-precision real-time localization is the guarantee of stable driving. Compared with the visual odometry (VO), the LiDAR odometry (LO) has the advantages of higher accuracy and better stability. However, 2D LO is only suitable for the indoor environment, and 3D LO has less efficiency in general. Both are not suitable for the online localization of an autonomous vehicle in an outdoor driving environment. In this paper, a direct LO method based on the 2.5D grid map is proposed. The fast semi-dense direct method proposed for VO is employed to register two 2.5D maps. Experiments show that this method is superior to both the 3D-NDT and LOAM in the outdoor environment.

Xudong He, Junqiao Zhao, Lu Sun et al.

In this paper, we studied a SLAM method for vector-based road structure mapping using multi-beam LiDAR. We propose to use the polyline as the primary mapping element instead of grid cell or point cloud, because the vector-based representation is precise and lightweight, and it can directly generate vector-based High-Definition (HD) driving map as demanded by autonomous driving systems. We explored: 1) the extraction and vectorization of road structures based on local probabilistic fusion. 2) the efficient vector-based matching between frames of road structures. 3) the loop closure and optimization based on the pose-graph. In this study, we took a specific road structure, the road boundary, as an example. We applied the proposed matching method in three different scenes and achieved the average absolute matching error of 0.07. We further applied the mapping system to the urban road with the length of 860 meters and achieved an average global accuracy of 0.466 m without the help of high precision GPS.

Yan Wu, Tao Yang, Junqiao Zhao et al.

The automatic parking is being massively developed by car manufacturers and providers. Until now, there are two problems with the automatic parking. First, there is no openly-available segmentation labels of parking slot on panoramic surround view (PSV) dataset. Second, how to detect parking slot and road structure robustly. Therefore, in this paper, we build up a public PSV dataset. At the same time, we proposed a highly fused convolutional network (HFCN) based segmentation method for parking slot and lane markings based on the PSV dataset. A surround-view image is made of four calibrated images captured from four fisheye cameras. We collect and label more than 4,200 surround view images for this task, which contain various illuminated scenes of different types of parking slots. A VH-HFCN network is proposed, which adopts an HFCN as the base, with an extra efficient VH-stage for better segmenting various markings. The VH-stage consists of two independent linear convolution paths with vertical and horizontal convolution kernels respectively. This modification enables the network to robustly and precisely extract linear features. We evaluated our model on the PSV dataset and the results showed outstanding performance in ground markings segmentation. Based on the segmented markings, parking slots and lanes are acquired by skeletonization, hough line transform and line arrangement.

Junqiao Zhao, Chen Ye, Yan Wu et al.

TiEV is an autonomous driving platform implemented by Tongji University of China. The vehicle is drive-by-wire and is fully powered by electricity. We devised the software system of TiEV from scratch, which is capable of driving the vehicle autonomously in urban paths as well as on fast express roads. We describe our whole system, especially novel modules of probabilistic perception fusion, incremental mapping, the 1st and the 2nd planning and the overall safety concern. TiEV finished 2016 and 2017 Intelligent Vehicle Future Challenge of China held at Changshu. We show our experiences on the development of autonomous vehicles and future trends.

Tao Yang, Yan Wu, Junqiao Zhao et al.

Semantic image segmentation is one of the most challenged tasks in computer vision. In this paper, we propose a highly fused convolutional network, which consists of three parts: feature downsampling, combined feature upsampling and multiple predictions. We adopt a strategy of multiple steps of upsampling and combined feature maps in pooling layers with its corresponding unpooling layers. Then we bring out multiple pre-outputs, each pre-output is generated from an unpooling layer by one-step upsampling. Finally, we concatenate these pre-outputs to get the final output. As a result, our proposed network makes highly use of the feature information by fusing and reusing feature maps. In addition, when training our model, we add multiple soft cost functions on pre-outputs and final outputs. In this way, we can reduce the loss reduction when the loss is back propagated. We evaluate our model on three major segmentation datasets: CamVid, PASCAL VOC and ADE20K. We achieve a state-of-the-art performance on CamVid dataset, as well as considerable improvements on PASCAL VOC dataset and ADE20K dataset