Terrain-aware Low Altitude Path Planning

2025-05-11

AI Summary

For terrain-aware navigation and obstacle avoidance in nap-of-the-earth (NOE) flight—relying solely on onboard RGB imagery and ego-pose—we propose an end-to-end vision-based navigation framework integrating behavior cloning (BC) and self-supervised learning. Our method jointly leverages self-supervised monocular depth estimation, motion-prior modeling, and a lightweight reinforcement learning policy network to generate collision-free trajectories in real time within a custom canyon simulation environment. The key innovation is the first-ever joint BC–self-supervised training paradigm, which markedly improves policy generalization under sparse expert supervision and robustness to complex, low-altitude terrain. Experimental results demonstrate a 23% increase in trajectory success rate, a 37% reduction in mean flight altitude, and inference latency under 50 ms—fully satisfying NOE real-time operational constraints.

Technology Category

• Search and Optimization: Learning to Search
• Computer Vision: Learning & Optimization for CV
• Humans and AI: Human-Aware Planning and Behavior Prediction

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Experiences and lessons learnt from Web-based algorithms and system deployments

Abstract

In this paper, we study the problem of generating low altitude path plans for nap-of-the-earth (NOE) flight in real time with only RGB images from onboard cameras and the vehicle pose. We propose a novel training method that combines behavior cloning and self-supervised learning that enables the learned policy to outperform the policy trained with standard behavior cloning approach on this task. Simulation studies are performed on a custom canyon terrain.

Problem

• Real-time low altitude path planning using RGB images
• Combining behavior cloning and self-supervised learning
• Improving policy performance for NOE flight simulation

Innovation

• Uses RGB images and vehicle pose
• Combines behavior cloning and self-supervised learning
• Simulated on custom canyon terrain