AGRNav: Efficient and Energy-Saving Autonomous Navigation for Air-Ground Robots in Occlusion-Prone Environments

TL;DR

AGRNav tackles safe and energy-efficient navigation for air-ground robots in occlusion-prone environments by predicting unseen obstacles and incorporating predictions into a low-latency occupancy map. The framework centers on SCONet, a lightweight semantic scene completion network that uses depthwise separable convolutions and two self-attention modules (CCA and MobileViT-v2) to predict occluded occupancy and semantics, paired with a query-based occupancy update and a hierarchical planner for energy-efficient trajectories. Key contributions include real-time SCONet performance on SemanticKITTI (IoU 56.12 and 20 FPS), a memory-efficient occupancy update reducing complexity to $O(M)$, and a validated energy-saving planner that reduces aerial path usage by about 50% in simulations and real-world tests. The results demonstrate safer, more energy-efficient navigation in occlusion-rich environments and provide open-source code for reproducibility.

Abstract

The exceptional mobility and long endurance of air-ground robots are raising interest in their usage to navigate complex environments (e.g., forests and large buildings). However, such environments often contain occluded and unknown regions, and without accurate prediction of unobserved obstacles, the movement of the air-ground robot often suffers a suboptimal trajectory under existing mapping-based and learning-based navigation methods. In this work, we present AGRNav, a novel framework designed to search for safe and energy-saving air-ground hybrid paths. AGRNav contains a lightweight semantic scene completion network (SCONet) with self-attention to enable accurate obstacle predictions by capturing contextual information and occlusion area features. The framework subsequently employs a query-based method for low-latency updates of prediction results to the grid map. Finally, based on the updated map, the hierarchical path planner efficiently searches for energy-saving paths for navigation. We validate AGRNav's performance through benchmarks in both simulated and real-world environments, demonstrating its superiority over classical and state-of-the-art methods. The open-source code is available at https://github.com/jmwang0117/AGRNav.

Figures (7)

• Figure 1: (a) Previous navigation systems had problems predicting occlusions, resulting in higher collision probabilities and suboptimal pathways that consumed more energy. (b) By predicting occlusions in advance, AGRNav can minimize and avoid collisions, resulting in efficient and energy-saving paths.
• Figure 2: The overview of our proposed Framework: AGRNav. $\mathbb{Q}$ denotes that the free voxels in the grid map query and update their occupancy status from the predicted occupancy map. $\mathbb{V}$ denotes that predicted semantics is turned into speed compensation.
• Figure 3: SCONet: Lightweight Semantic Scene Completion Network. Our network employs a self-attention-driven U-Net architecture, featuring depthwise separable convolutions and segmentation heads, to perform efficient 3D scene completion and semantic segmentation.
• Figure 4: Four methods were used to plan paths in a simulated square room. AGRNav demonstrates the ability to predict the distribution of obstacles in occluded areas.
• Figure 5: The detailed composition of our customized air-ground robot (AGR).