Autonomous exploration is a fundamental problem for certain applications of unmanned ground vehicles (UGVs). Depth cameras are widely used in exploring indoor and specific outdoor environments. However, existing methods suffer from inefficient and unstable performance of exploration in open outdoor environments. This study develops a framework to achieve fast and robust three-dimensional exploration by UGVs with a gimbal camera in challenging outdoor environments. A hierarchical reactive planning approach based on a multilayered map is proposed to improve exploration efficiency. Additionally, an adaptive semantic mapping algorithm is proposed to more accurately and stably represent the environmental localization uncertainty. Localization uncertainty is integrated into the planning approach to decrease odometry drift and prevent localization failure, further enhancing the exploration robustness. A viewpoint tracking controller based on terrain-aware model predictive control is proposed to guarantee the accuracy of the planned viewpoint tracking and to further improve the robustness of localization and exploration on rough terrains. Finally, several physical-engine simulations and experiments in outdoor environments are performed. The comparisons to existing state-of-the-art approaches have verified the effectiveness of the proposed approach. The code of our system will be available at https://github.com/HITSZ-NRSL/Open3DExplorer.git.

中文翻译：

---

带有万向节 RGB-D 相机的移动机器人的全身运动规划和跟踪，用于户外 3D 探索

自主探索是无人地面车辆（UGV）某些应用的一个基本问题。深度相机广泛用于探索室内和特定的室外环境。然而，现有方法在开放的室外环境中探索效率低且不稳定。本研究开发了一个框架，可在具有挑战性的室外环境中使用带有万向摄像头的无人车实现快速、稳健的三维探索。提出了一种基于多层地图的分层反应规划方法以提高勘探效率。此外，提出了一种自适应语义映射算法，以更准确、稳定地表示环境定位的不确定性。定位不确定性被集成到规划方法中，以减少里程漂移并防止定位失败，进一步增强探索的鲁棒性。提出了一种基于地形感知模型预测控制的视点跟踪控制器，以保证规划视点跟踪的准确性，并进一步提高崎岖地形上定位和探索的鲁棒性。最后，在室外环境中进行了多次物理引擎模拟和实验。与现有最先进方法的比较验证了所提出方法的有效性。我们系统的代码可在 https://github.com/HITSZ-NRSL/Open3DExplorer.git 获取。