RAMA: Retrieval-Augmented Multi-Agent Routing with Episodic Memory for LEO Satellite Networks
What it is
RAMA is a deep reinforcement learning framework for routing in Low-Earth Orbit satellite networks that integrates local observations, routing history, and retrieved episodic memories using a Transformer-based decision network. It leverages Approximate Information State (AIS) to improve routing decisions, significantly reducing end-to-end latency compared to standard DDQN methods in large-scale constellations.
The problem it solves
Existing deep reinforcement learning routing methods in LEO networks fail to exploit successful past experiences, leading to suboptimal routing decisions, long detours, and performance degradation as network conditions change.
What you could use it for
- Optimizing packet routing in large-scale LEO satellite constellations like Starlink
- Reducing tail latency in dynamic satellite network topologies
- Implementing decentralized routing agents that learn from past successful paths
Why it matters
Significantly improves the efficiency and stability of data transmission in next-generation satellite internet services by reducing latency and avoiding network congestion through learned experience retrieval.
Read the source ↗Back to the library →
Summary written by an AI model from the source; check the source before you rely on it.