Edge devices capture more than any link can carry. Mission Link sends only what the receiver's mission needs — a fixed 81-byte message that outperforms sending the raw feed at the same bandwidth.
A single drone generates hundreds of KB to tens of MB every second, dominated by video. The tactical link is narrow, jammed, and often relayed across several hops. You cannot send it all — so most systems either choke the link or throw the mission-critical detail away. Mission Link sends the meaning, not the pixels.
Mission Link is a task-conditioned semantic-compression middleware. It encodes each sensor scene into a fixed message with two channels — structured facts in a schema, un-schematizable detail in a neural latent. The receiver's objective decides what each message keeps.
Neither channel alone is enough — structure gets the position, the neural latent gets the meaning. Together, in 81 bytes, they match sending the full feed.
In our experiments, learned semantic compression dominates the rate–distortion frontier: at every matched transmission rate, it beats sending a downsampled raw feed.
81 bytes sits in the middle of real tactical data links — fixed and deterministic, so it slots into existing TDMA schedules and bandwidth budgets.
Send meaning over the robust omnidirectional link; escalate to a high-bandwidth line-of-sight link only when actual imagery is truly needed.
Fuse many drones' 81-byte messages to coordinate; when a peer goes dark, a shared model predicts what it will do next.
Light enough to run on the airframe in real time; the structured channel is a human-readable field a person can read and verify.
Arvenne builds Mission Link — task-aware semantic communication for unmanned and edge systems. We are developing it for Korea's 2026 defense-technology startup competition.