Real-time Multi-Plane Segmentation Based on GPU Accelerated High-Resolution 3D Voxel Mapping for Legged Robot Locomotion

TL;DR

This work tackles the need for real-time, high-fidelity perception for legged locomotion by extending 3D voxel mapping with GPU-accelerated multi-plane segmentation. The authors introduce a framework that maintains a $3$D voxel map at $0.01$ m resolution in a robot-centric frame and applies vertex-based CCL clustering, cluster-parallel RANSAC plane estimation, and GPU-convex hull boundary generation to produce polygonal planes. They demonstrate real-time performance (over $30\ \mathrm{Hz}$) and high plane IoU on both simulated and physical legged robots, enabling safe locomotion in environments with open-tread stairs and overhead structures. The approach supports small obstacle detection, RL-based safety enhancements, and is validated across diverse sensors and platforms, highlighting practical impact for robust, geometry-aware locomotion.

Abstract

This paper proposes a real-time multi-plane segmentation method based on GPU-accelerated high-resolution 3D voxel mapping for legged robot locomotion. Existing online planar mapping approaches struggle to balance accuracy and computational efficiency: direct depth image segmentation from specific sensors suffers from poor temporal integration, height map-based methods cannot represent complex 3D structures like overhangs, and voxel-based plane segmentation remains unexplored for real-time applications. To address these limitations, we develop a novel framework that integrates vertex-based connected component labeling with random sample consensus based plane detection and convex hull, leveraging GPU parallel computing to rapidly extract planar regions from point clouds accumulated in high-resolution 3D voxel maps. Experimental results demonstrate that the proposed method achieves fast and accurate 3D multi-plane segmentation at over 30 Hz update rate even at a resolution of 0.01 m, enabling the detected planes to be utilized in real time for locomotion tasks. Furthermore, we validate the effectiveness of our approach through experiments in both simulated environments and physical legged robot platforms, confirming robust locomotion performance when considering 3D planar structures.

Figures (12)

• Figure 1: The proposed method enables real-time high-precision 3D multi-plane segmentation for legged robot locomotion (a), detects 3D multi-layered planar surfaces (b), and enables 3D locomotion under structures such as desks (c).
• Figure 2: Framework overview of the proposed method. The framework consists of a mapping module and a multi-plane segmentation module. The mapping module accumulates point clouds in a 3D voxel map and removes dynamic objects through ray casting operations. The multi-plane segmentation module classifies accumulated points into steppable and object points, clusters the steppable points, and detects multiple planes from the clustered steppable points.
• Figure 3: 3D voxel mapping module: The voxel update module accumulates point clouds in the 3D voxel map, while the voxel clearing module removes dynamic objects by casting rays from the sensor to each point cloud location.
• Figure 4: Simulation environments used for evaluation.
• Figure 5: Processing time analysis in the five-step stair environment: (a) Mapping module processing time comparison across different sensor configurations. The proposed GPU-based 3D voxel mapping method achieves sufficiently high-speed mapping execution. (b) Multi-plane segmentation module processing time comparison. The proposed GPU-based multi-plane segmentation method maintains significantly superior performance compared to existing CPU-based methods.