Where Do We Go from Here?
Charting the Future of Unmanned Vehicles

01 · Motivation

A booming field, growing in three separate directions

Unmanned vehicles operate without a human onboard — remotely piloted or fully autonomous. They come in three flavours, one per element: UAV — air UGV — ground USV — sea. The money is real and accelerating — drag the year:

Each platform already earns its keep across four sectors. Explore the application landscape from the paper — click any cell:

Here is the catch: research on each platform optimizes that platform. UAV work makes UAVs more efficient; USV work makes USVs more capable. But real missions — military, manufacturing, disaster response — almost never involve just one category. The platforms can complement, compound, or outright contradict each other. Operational efficiency in isolation is the wrong objective.

02 · The problem

Watch the silo fail

Take the paper's first scenario: a forest wildfire, goal: find and evacuate everyone. UGVs can carry supplies and escort people — but from the ground, finding them is brutally slow. A UAV swarm sees the whole forest from above — but can't carry anyone. Below, the same fire, the same trapped civilians, the same vehicles. The only difference: on the left the domains don't talk; on the right the UAVs feed every sighting to the UGVs. Press play and let them race.

The second scenario is maritime security: neutralize a hostile vessel before it reaches the harbour. A USV alone can do it — sonar works — but every minute of late detection is a minute closer to the harbour. Step through what each extra domain buys:

These failures look like communication problems. They are not. The vehicles could exchange radio packets all day — the real bottleneck is the fusion model: nothing in today's stacks can merge what a sonar ping, an aerial image and a ground patrol mean for the mission. The paper names the missing capability Cross-domain Mission Intelligence.

03 · The core idea

Information fusion is a ladder — and everyone is stuck on rung two

Fusion is not one monolithic process; it is a hierarchy of increasing abstraction. Each level answers a bigger question than the last. Click the levels of the pyramid — each comes with a live demonstration of what that level actually computes:

Rung one and rung two are well-studied: Kalman filters, SLAM, collaborative multi-object tracking, covariance intersection. But they are syntactic and geometric — they answer “what” and “where”, never “why” or “so what?”. Mission-level fusion trades states for intent, capability and causality: resolving competing objectives, reasoning about missing capabilities, predicting cross-domain impacts. That leap needs new primitives — structured world models that capture dependencies among agents, tasks and environments. Building toward it is the core of this paper.

04 · Prior work

Everyone solved a fragment

The dream of cross-domain swarms is not new. But every prior attempt solved one slice and stopped. This is the paper's comparison table made explorable — click a row to see exactly where each approach hits its ceiling, as a working sketch:

Simplified multi-swarm partitioning, syntactic messaging standards, abstract game theory, task-specific physical coordination — fragments. What is missing is a holistic architecture that lets heterogeneous swarms reason collectively about multi-objective missions. That is the gap the UxV Ecosystem Architecture aims at.

05 · The proposal

The UxV Ecosystem Architecture

Four layers, three design principles: layered abstraction to separate concerns, decentralization so nothing is a single point of failure, and service-oriented design for plug-and-play modularity. Every box below is clickable. Then press the flow buttons to watch a mission travel down the stack — and sensor reality travel back up.

Four working assumptions keep the blueprint honest: every vehicle keeps enough local intelligence to stay safe on its own; every vehicle is swarm-enabled with a Gateway to the ecosystem; connectivity may be disturbed but not permanently severed; and the assigned mission is within the fleet's collective capability. The architecture's job is to unlock that capability, not to conjure it.

06 · Mechanisms, animated

Four moving parts that make it work

6.1  Delay-tolerant messaging: no connection, no problem

The communication bus is publish–subscribe, so producers and consumers never couple directly — and it speaks Delay-Tolerant Networking: store-and-forward messaging for communication-denied environments. Watch a UAV fly beyond radio range: its observations buffer on board, and the moment it reconnects, everything syncs into the shared world model. Shrink the radio range and try to break it.

6.2  Service directory: the fleet's yellow pages

Capabilities live in a distributed hash table: when a vehicle joins, its Gateway registers what it can do against a service ontology. Any vehicle can then discover, at mission time, who can fulfil a need it cannot. Step through the paper's example — a UGV inspecting a power plant stumbles on a chemical spill:

6.3  Spatio-temporal fusion: two mediocre sensors, one sharp track

Inside the Distributed World Model, the fusion engine performs data association (are these two detections the same boat?) and state fusion (merge them into one track). Each sensor alone is noisy in its own way — radar is precise in range but sloppy in bearing; a camera is the opposite. Toggle them and watch the fused uncertainty collapse below either sensor's own:

6.4  Task allocation: the market decides, then re-decides

The Collaborative Task Engine decomposes a mission (Evacuate Area → search grid A, guide to exit 1, …) and auctions each task: vehicles bid their estimated cost in energy and time. A task needing aerial imagery draws cheap bids from UAVs and absurd ones from UGVs — allocation happens naturally, no central planner. And when a winner dies mid-mission, the engine simply re-runs the auction. Try it:

07 · Put it all together

Run the ecosystem yourself

Everything above, in one sandbox: the wildfire mission with the full ecosystem loop — UAVs scout and publish to the world model, the task engine auctions every discovered civilian to the cheapest UGV, the USV ferries anyone rescued near the river, and re-tasking is continuous. Then sabotage it: jam communications (watch DTN buffers grow), or click any vehicle to kill it (watch its tasks re-auction). Flip back to siloed mode to remind yourself what all this machinery buys.

08 · Challenges & roadmap

What stands between blueprint and reality

And five directions the paper calls the community toward — click through:

09 · Takeaways

What this paper changes

10 · Resources

Paper & citation

• Paper: Where Do We Go from Here? Charting the Future of Unmanned Vehicles — IEEE Xplore (doi:10.23919/WMNC67099.2025.11299247)
• Open access PDF: IFIP Open Digital Library
• Venue: 16th IFIP Wireless and Mobile Networking Conference (WMNC 2025), Leuven, Belgium, pp. 176–181.
• Funding: part of the STEADFAST project (NWO KiC programme), partly financed by the Dutch Research Council (NWO).

@inproceedings{latha2025unmanned,
  author    = {Nanjaiya Latha, Adarsh and Anbalagan, Sabari Nathan and Chiumento, Alessandro and Laarhuis, Jan},
  title     = {Where Do We Go from Here? Charting the Future of Unmanned Vehicles},
  booktitle = {Proc. 16th IFIP Wireless and Mobile Networking Conference (WMNC)},
  address   = {Leuven, Belgium},
  pages     = {176--181},
  year      = {2025},
  doi       = {10.23919/WMNC67099.2025.11299247}
}

The animations on this page are conceptual reconstructions built to teach the paper's ideas; scenario dynamics are illustrative, not experimental data.