LEARN: Learning End-to-End Aerial Resource-Constrained Multi-Robot Navigation

2025-11-21

AI Summary

Autonomous navigation of resource-constrained nano-drone swarms in complex environments remains challenging due to severe computational, power, and sensing limitations. Method: This paper proposes a two-stage safety-guided lightweight reinforcement learning (RL) framework integrating low-resolution time-of-flight (ToF) sensing, an attention-driven lightweight RL policy, a simple motion planner, and a distributed coordination algorithm—enabling fully onboard, low-power multi-agent formation navigation. Contribution/Results: To the best of our knowledge, this is the first work to achieve fully autonomous indoor and outdoor flight of six Crazyflie nano-drones without external infrastructure. The system supports speeds up to 2.0 m/s and navigates through gaps as narrow as 0.2 m. In simulation, it outperforms state-of-the-art methods by 10% in navigation performance, empirically validating the feasibility and effectiveness of joint lightweight design across perception, decision-making, and communication modules.

Technology Category

• Natural Language Processing: Safety and Robustness
• Intelligent Robots: Localization, Mapping, and Navigation
• Search and Optimization: Learning to Search

Application Category

• Responsible Web: Machine-in-the-loop, human agency and autonomy
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Security and Privacy: Large-scale security measurements

Abstract

Nano-UAV teams offer great agility yet face severe navigation challenges due to constrained onboard sensing, communication, and computation. Existing approaches rely on high-resolution vision or compute-intensive planners, rendering them infeasible for these platforms. We introduce LEARN, a lightweight, two-stage safety-guided reinforcement learning (RL) framework for multi-UAV navigation in cluttered spaces. Our system combines low-resolution Time-of-Flight (ToF) sensors and a simple motion planner with a compact, attention-based RL policy. In simulation, LEARN outperforms two state-of-the-art planners by $10%$ while using substantially fewer resources. We demonstrate LEARN's viability on six Crazyflie quadrotors, achieving fully onboard flight in diverse indoor and outdoor environments at speeds up to $2.0 m/s$ and traversing $0.2 m$ gaps.

Problem

• Addresses nano-UAV navigation challenges from constrained sensing and computation
• Develops lightweight RL framework for multi-robot navigation in cluttered spaces
• Enables fully onboard flight using low-resolution sensors and compact policies

Innovation

• Lightweight two-stage safety-guided RL framework
• Combines low-resolution ToF sensors with attention-based policy
• Uses simple motion planner for resource-constrained navigation