In this paper, we present Distributed Brain, a novel Distributed JEPA-Assisted Semantic Communication System designed for "Intuitive" Sensing. Traditional distributed communication systems primarily rely on raw-data transmission, which is computationally expensive and bandwidth-intensive. In contrast, our approach leverages the Joint Embedding Predictive Architecture (JEPA) to extract and transmit compact, meaningful embeddings instead of raw data. This reduces communication overhead, improves data privacy, and enables efficient cross-modal sensing. We experimentally demonstrate that our example system, composed of server, cameras, and CSI sensor nodes, collaboratively performs complex tasks that no single node can achieve independently. Our experiments show that transmitting embeddings significantly reduces data transmission volume and communication latency while maintaining sensing accuracy. Furthermore, we discuss the trade-offs between modality count and model complexity, and highlight how this system opens a new avenue for intuitive and proactive sensing in distributed environments.

We experimentally demonstrate that our example system, composed of a server, cameras, and CSI sensor nodes, collaboratively performs complex tasks that no single node can achieve independently. Our experiments show that transmitting embeddings significantly reduces data transmission volume and communication latency while maintaining sensing accuracy. Furthermore, we discuss the trade-offs between modality count and model complexity, and highlight how this system opens a new avenue for intuitive and proactive sensing in distributed environments.

In this paper, we propose JEPA-ISAC, a JEPA-aided Multimodal Integrated Sensing and Communication framework to advance multimodal semantic sensing and fusion within the Integrated Sensing and Communication (ISAC) paradigm.

By leveraging the Joint Embedding Predictive Architecture (JEPA), we enable cross-modal semantic integration between wireless communication signals and other sensory data such as images and LiDAR.

leverages the Joint Embedding Predictive Architecture (JEPA) to extract and transmit compact, meaningful embeddings instead of raw data.

This reduces communication overhead, improves data privacy, and enables efficient cross-modal sensing.