SLAMs for Quadruped Robot: Mapping Capability-centric Agentic Simulation

The integration of multi-sensors, particularly LiDAR, into quadruped robots is crucial for enhancing their mapping capabilities. However, the unique motion dynamics of these robots pose significant integration challenges. To address this, we introduce QR-SimEval, a specialized simulation intelligent agent for evaluating 3D SLAM algorithms on quadruped platforms. QR-SimEval supports nine SLAM algorithms (including our proposed C-MAPS) across 15 diverse simulated environments, from indoor labs to complex outdoor terrains. Comprehensive comparisons evaluate mapping performance and odometry accuracy. Notably, experiments using approximately 571 million spatial 3D points from 15 challenging real-world scenarios demonstrate that quadruped robots optimized via QR-SimEval achieve exceptional 3D point cloud map construction, validating the effectiveness of our simulation architecture in improving mapping perception capabilities. This work provides crucial guidance for robust sensor integration and algorithm deployment in quadruped robots.