CoNav-UAV: Stackelberg Learning for Dual-Altitude Aerial Navigation
A recent study presents CoNav-UAV, a framework designed for vision-and-language navigation (VLN) focused on targets for aerial vehicles. This research tackles difficulties faced in operations like disaster relief, infrastructure evaluation, and security monitoring, where an unmanned aerial vehicle (UAV) must identify targets using a brief description of their features and environment. The challenge involves both global exploration and collision-free close-range navigation, which are complex to integrate within a single agent. Existing approaches often adapt ground VLN techniques for low-altitude UAVs, relying on external support for exploration. A novel method utilizes two UAVs at different heights but depends on privileged data and trains the agents separately, hindering necessary cooperation. CoNav-UAV frames the task as a Stackelberg game between a high-altitude and a low-altitude UAV, facilitating collaborative navigation. The paper can be found on arXiv with the identifier 2608.01802.
Key facts
- CoNav-UAV is a framework for cooperative dual-altitude aerial navigation.
- It uses Stackelberg learning to model the interaction between a high-altitude and a low-altitude UAV.
- The task is target-oriented vision-and-language navigation (VLN) on aerial platforms.
- Applications include disaster rescue, infrastructure inspection, and security patrol.
- Existing methods transfer ground VLN to low-altitude UAVs with external assistance.
- A recent attempt uses two UAVs at complementary altitudes but relies on privileged information and independent training.
- CoNav-UAV addresses the need for mutual adaptation between agents.
- The paper is published on arXiv with ID 2608.01802.
Entities
Institutions
- arXiv