Latent Telepathy: Multi-Robot Communication with Self-Supervised Perceptual Latents
Latent Telepathy: Multi-Robot Communication with Self-Supervised Perceptual Latents
The paper presents a method for multi-robot communication in decentralized teams under partial observability. By broadcasting perceptual latent vectors derived from a self-supervised encoder, robots can share compressed perceptual information without explicit task coupling. The approach achieves high reliability in navigation tasks, outperforming traditional kinematic messages and raw images, and is validated across multiple environments.
This paper introduces 'Latent Telepathy,' a novel approach to multi-robot communication that leverages self-supervised perceptual latents to enable decentralized decision-making. The method allows robots to share compressed perceptual information without explicit task coupling, achieving high reliability in navigation tasks while maintaining low computational overhead.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
Source：arXiv Robotics — research abstracts · arxiv.org