Miles Michael J, Biggie Harel, Heckman Christoffer
Department of Mechanical Engineering, University of Colorado Boulder, Boulder, CO, United States.
Department of Computer Science, University of Colorado Boulder, Boulder, CO, United States.
Front Robot AI. 2023 Oct 3;10:1249586. doi: 10.3389/frobt.2023.1249586. eCollection 2023.
Navigation over torturous terrain such as those in natural subterranean environments presents a significant challenge to field robots. The diversity of hazards, from large boulders to muddy or even partially submerged Earth, eludes complete definition. The challenge is amplified if the presence and nature of these hazards must be shared among multiple agents that are operating in the same space. Furthermore, highly efficient mapping and robust navigation solutions are absolutely critical to operations such as semi-autonomous search and rescue. We propose an efficient and modular framework for semantic grid mapping of subterranean environments. Our approach encodes occupancy and traversability information, as well as the presence of stairways, into a grid map that is distributed amongst a robot fleet despite bandwidth constraints. We demonstrate that the mapping method enables safe and enduring exploration of subterranean environments. The performance of the system is showcased in high-fidelity simulations, physical experiments, and Team MARBLE's entry in the DARPA Subterranean Challenge which received third place.
在诸如天然地下环境中的崎岖地形上导航，对野外机器人来说是一项重大挑战。从大石块到泥泞甚至部分被淹没的地面，各种危险层出不穷，难以完全界定。如果这些危险的存在和性质必须在同一空间中运行的多个智能体之间共享，挑战就会加剧。此外，高效的地图绘制和强大的导航解决方案对于半自主搜索和救援等行动绝对至关重要。我们提出了一个用于地下环境语义网格地图绘制的高效且模块化的框架。我们的方法将占用和可通行性信息以及楼梯的存在编码到一个网格地图中，尽管存在带宽限制，但该地图仍在机器人机群中进行分发。我们证明了这种地图绘制方法能够实现对地下环境的安全且持久的探索。该系统的性能在高保真模拟、物理实验以及团队MARBLE参加的获得第三名的DARPA地下挑战赛中得到了展示。
