Visual-Geometry GP-based Navigable Space for Autonomous Navigation

TL;DR

The simulation and real-world experiments demonstrate that the proposed VG-SGP model, coupled with the innovative navigation strategy, outperforms models solely reliant on visual or geometric navigation algorithms, highlighting a superior adaptive behavior.

Abstract

Autonomous navigation in unknown environments is challenging and demands the consideration of both geometric and semantic information in order to parse the navigability of the environment. In this work, we propose a novel space modeling framework, Visual-Geometry Sparse Gaussian Process (VG-SGP), that simultaneously considers semantics and geometry of the scene. Our proposed approach can overcome the limitation of visual planners that fail to recognize geometry associated with the semantic and the geometric planners that completely overlook the semantic information which is very critical in real-world navigation. The proposed method leverages dual Sparse Gaussian Processes in an integrated manner; the first is trained to forecast geometrically navigable spaces while the second predicts the semantically navigable areas. This integrated model is able to pinpoint the overlapping (geometric and semantic) navigable space. The simulation and real-world experiments demonstrate that the ability of the proposed VG-SGP model, coupled with our innovative navigation strategy, outperforms models solely reliant on visual or geometric navigation algorithms, highlighting a superior adaptive behavior.

Figures (8)

• Figure 1: Visual-Geometry Navigable Space: (a) shows LiDAR's pointcloud (yellow) and camera's colored-pointcloud; (b) shows the geometry navigable space (grey-coded circular surface, white regions represent free space), the visual navigable space (grey-coded vertical plane, white regions represents the navigable classes in the camera field of view), and the local navigation points (colored-circles, where color represents the go-to-goal cost).
• Figure 2: Geometry-Navigable Space: (b) shows the raw pointcloud in yellow and the original occupancy surface , where warmer colors indicate less occupancy; (c) shows the predicted occupancy surface using the G-SGP model; (d) shows the variance surface, i.e., the uncertainty associated with the predicted occupancy, where white color indicates highly uncertain (free) points. The Geometry-LNPs (G-LNPs) are shown as colored circles, where green indicates less go-to-goal cost associated with each G-LNP.
• Figure 3: Visual-Navigable Space: V-LNPs are shown as colored-circles in (h), where the color shows the cost assigned to each V-LNP.
• Figure 4: D-SGP and D-SGP models of camera's pointcloud: both (a) and (b) shows the pointcloud generated from the D-SGP model, wherein the grey color indicates the uncertainty in (a) and the predicted navigability class in (b).
• Figure 5: Simulation Experiments: VG-LNPs cost on the right figure. $K_{nav}>0.5$ giving less cost for visually VG-LNPs than G-LNPS outside the Camer'a FoV.