Scholarlink

<p>Robots that can perceive and map the world in three dimensions are becoming essential to support exploration, environmental monitoring and infrastructure inspection, yet their performance in complex environments is often limited by motion and sensing constraints. Quadruped robots can navigate uneven, cluttered and confined spaces, but their highly dynamic gait disrupts conventional light detection and ranging-based simultaneous localization and mapping (SLAM) methods, creating a gap between algorithm design and real-world deployment. Here we present an agentic simulation framework, Quadruped Robot Motion and Mapping Simulation Evaluation Agent (QR-SimEval), together with an integrated multisensor mapping system, C-MAPS, to enable the realistic evaluation and optimization of three-dimensional SLAM on quadruped robots. QR-SimEval supports nine SLAM algorithms (including C-MAPS) across 15 simulated environments, from indoor laboratories to complex outdoor terrains, reproducing diverse motions and structures while enabling comprehensive benchmarking. C-MAPS achieves stable, drift-free mapping in both simulated and real challenging terrains. These results demonstrate that agent-driven simulation can improve simulation-to-reality transfer, advancing robust three-dimensional mapping in unstructured environments.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Mapping capability-centric agentic simulations for SLAMs on quadruped robots

  • 期刊论文

  • 发表日期：

    2026年4月1日

查看全文
  • Yanyi Li,
  • Akram Akbar,
  • Chun Liu,
  • Yunze Wu,
  • Honghao Wang,
  • Jing Guo,
  • Zhijian Guo

摘要

Robots that can perceive and map the world in three dimensions are becoming essential to support exploration, environmental monitoring and infrastructure inspection, yet their performance in complex environments is often limited by motion and sensing constraints. Quadruped robots can navigate uneven, cluttered and confined spaces, but their highly dynamic gait disrupts conventional light detection and ranging-based simultaneous localization and mapping (SLAM) methods, creating a gap between algorithm design and real-world deployment. Here we present an agentic simulation framework, Quadruped Robot Motion and Mapping Simulation Evaluation Agent (QR-SimEval), together with an integrated multisensor mapping system, C-MAPS, to enable the realistic evaluation and optimization of three-dimensional SLAM on quadruped robots. QR-SimEval supports nine SLAM algorithms (including C-MAPS) across 15 simulated environments, from indoor laboratories to complex outdoor terrains, reproducing diverse motions and structures while enabling comprehensive benchmarking. C-MAPS achieves stable, drift-free mapping in both simulated and real challenging terrains. These results demonstrate that agent-driven simulation can improve simulation-to-reality transfer, advancing robust three-dimensional mapping in unstructured environments.