metric obstacle mapping

LiDAR-based localization suffers significant accuracy degradation in feature-deprived environments (e.g., long straight corridors, blank walls). To address this, this paper proposes a geometry-aware trajectory planning framework. Our method introduces: (1) a Geometric Feature Metric (GFM) that quantifies local environment observability for robust pose estimation; (2) a Mesh-encoded Metric (MEM) map enabling constant-time pose decoding and real-time feature awareness; and (3) integration of GFM into trajectory optimization to explicitly maximize localization robustness during path planning. Extensive simulations and real-world experiments demonstrate that the proposed approach reduces localization error by 37.2% in feature-sparse scenarios, substantially improving navigation reliability and system robustness. To the best of our knowledge, this is the first work achieving closed-loop co-optimization of perception quality, localization performance, and motion planning.

To address the challenge of rapidly detecting path infeasibility in robot motion planning, this paper proposes the first lightweight discriminative framework based on incremental sampling and image segmentation. Methodologically, it discretizes the configuration space, performs obstacle-guided incremental sampling to construct a binary occupancy map, and then applies connected-component analysis—inspired by image segmentation—to determine whether the start and goal configurations reside in the same free-connected component. Crucially, the framework avoids full configuration-space modeling and eliminates dependence on path-search algorithms. Evaluated across five representative scenarios with up to 5 degrees of freedom, it achieves high-accuracy infeasibility identification while reducing average detection time by approximately two orders of magnitude compared to conventional feasibility verification methods such as RRT and PRM, thereby significantly lowering computational overhead.

To address insufficient environmental perception accuracy and challenges in dynamic collision modeling and real-time detection for industrial robot motion planning, this paper proposes a multi-sensor fusion approach for high-precision workspace registration and collision detection. By fusing point clouds from depth cameras and LiDAR, we design a registration algorithm integrating region-growing segmentation with VCCS-based supervoxel clustering, significantly improving the accuracy and robustness of obstacle identification in complex industrial scenes. Furthermore, we employ point-cloud approximation modeling coupled with an optimized 3D collision detection algorithm to construct a lightweight, dynamic, and incrementally updatable environmental model. Experimental results demonstrate that the method achieves millisecond-level response latency while reducing modeling error by 32% and attaining a collision detection accuracy of 99.1%. The framework has been successfully integrated into a real-world industrial robot system, validating its effectiveness for real-time obstacle avoidance.

本文解决了约束运动规划问题，通过引入诱导黎曼度量来统一不同表示方法下的路径长度计算，从而在采样规划和轨迹优化中实现一致的几何处理。

To address cumulative pose drift in visual-inertial odometry (VIO/SLAM), which causes erroneous obstacle classification and compromises navigation safety, this paper proposes the first verifiable certified mapping framework. Our method models incremental pose uncertainty propagation geometrically, inflates safe regions accordingly, and formally proves the correctness of both Safe Flight Corridors and Signed Distance Fields under bounded pose error. By tightly coupling uncertainty-aware pose estimation, geometric safety inflation, and trajectory planning/control, the framework guarantees strict trustworthiness of obstacle-free regions in the map. Evaluated on the Replica dataset, our approach significantly outperforms state-of-the-art methods. In real-robot experiments, it achieves zero collisions—whereas all baseline methods incur at least one collision—demonstrating both theoretical soundness and practical robustness for safety-critical autonomous navigation.

This work addresses the lack of Riemannian metric-driven obstacle avoidance in existing motion planning libraries by proposing an open-source C++20 framework that, for the first time, employs configuration-dependent Riemannian metrics as core geometric drivers for distance computation and interpolation. By decoupling components such as manifolds and metrics, the library provides a unified sampling-based planning interface that enables seamless transitions across multiple spaces, while integrating Lie group geometric primitives and Python bindings. Experimental evaluations demonstrate that the proposed approach generates shorter, more energy-efficient trajectories across diverse manifolds. The project is accompanied by comprehensive documentation, extensive testing, and a fully reproducible benchmark suite.

This work addresses the challenge that monocular vision systems struggle to simultaneously achieve globally consistent localization and metrically accurate dense obstacle perception, while multi-sensor approaches suffer from complex calibration and high costs. The authors propose a unified framework that relies solely on monocular RGB input, innovatively leveraging ground geometry to provide online scale constraints that effectively resolve monocular scale ambiguity. By fusing pose-anchored visual geometry with physical scale priors, the method jointly optimizes metric localization and obstacle perception. It produces a metrically consistent obstacle representation directly usable for path planning, demonstrates strong generalization across diverse environments, and has been successfully deployed on real mobile robots, validating its practical utility for low-cost, scalable, and safe autonomous navigation.

为解决几何感知与机器人动作对齐问题，提出Metric-Aware Geometry Perception框架，通过相机参数和深度观测重建度量几何，提高跨场景一致性。

This work addresses the high communication overhead and low data efficiency in multi-robot collaborative SLAM, which often stems from reliance on low-level feature matching. The authors propose a distributed SLAM framework based on scene graph matching that leverages RGB-LiDAR fused point clouds for semantic segmentation, extracting discrete objects and their boundaries to construct scene graphs. Notably, inter-robot loop closures are achieved solely by exchanging object labels and centroids, eliminating dependence on raw feature descriptors. Integrated with a multi-stage communication scheme and distributed optimization, the method significantly reduces communication load while preserving localization and mapping accuracy, as demonstrated in both simulated and real-world experiments with legged robots across indoor and outdoor environments.

研究提出DCLP++框架，利用足迹清除和相对运动特性解决动态环境中的局部导航问题，通过LiDAR数据映射及策略学习提高导航成功率。