Drone swarms have a communication problem, and the solution may be light. Researchers at Worcester Polytechnic Institute (WPI), a technical university in Worcester, Massachusetts, and MIT Lincoln Laboratory, a federally funded research and development center operated by MIT for the U.S.
image from FLUX 2.0 Pro
Drone swarms have a communication problem, and the solution may be light.
Researchers at Worcester Polytechnic Institute (WPI), a technical university in Worcester, Massachusetts, and MIT Lincoln Laboratory, a federally funded research and development center operated by MIT for the U.S. Air Force, have built an optical communication system that lets robots talk to each other using flashing LEDs and event cameras — no radio waves, no jamming risk. A preprint posted to arXiv on March 19 describes a system that achieves over 95 percent decoding accuracy while tracking a moving LED transmitter at frequencies of at least 1 kilohertz. The processing is seven times faster than the previous state of the art.
The work is Air Force funded. Two of the six authors — Brendan Long and Matthew Cleaveland — are researchers at MIT Lincoln Laboratory, which operates under Air Force contracts administered at Hanscom Air Force Base in Massachusetts. The paper lists two contract numbers: FA8702-15-D-0001 and FA8702-25-D-B002. The copyright is held by MIT and delivered to the U.S. Government with Unlimited Rights under DFARS — the standard language for DoD-owned research cleared for public release.
What the Air Force is paying for, in plain terms: drone swarms that can coordinate even when someone is jamming their radios.
The operational urgency is not hypothetical. IEEE Spectrum reported that tens of thousands of jammers have been deployed on Ukrainian front lines, forcing drone manufacturers to scramble for alternatives. KrattWorks, an Estonian drone company, adopted optical neural-net navigation for its Ghost Dragon quadcopter specifically to overcome GPS and RF jamming. The cat-and-mouse game, KrattWorks COO Martin Karmin said, is moving extremely fast. The Associated Press confirmed that both Russia and Ukraine have deployed fiber-optic tethered drones as a workaround — the tether makes jamming irrelevant. But a tethered drone is a solo drone. Tethers kill the dynamics that make swarms worth building. A RAND Corporation commentary from November 2025 framed electromagnetic warfare as NATO's blind spot: control over the invisible battlespace where communications are jammed and drones are blinded can decide the outcome of a conflict.
The WPI/MIT LL system is designed for the untethered case. Instead of radio frequencies, robots use flashing LEDs as transmitters and event cameras as receivers. Event cameras — unlike conventional frame-based cameras — don't capture video at a fixed frame rate. They register individual pixel-level changes as they happen, with microsecond resolution, and they handle high dynamic range environments that would blind a normal camera. That makes them extraordinarily sensitive to a blinking light, even when both the camera and the light source are in motion.
The gap the new paper closes is a real one. Prior work by Ziwei Wang and colleagues, published as a preprint to arXiv in 2022 and presented at IROS, showed event cameras could receive LED-encoded messages at 4 kilobits per second indoors and 500 bits per second at 100 meters outdoors — but only between a stationary transmitter and stationary receiver. A 2025 preprint by Hang Su and colleagues, later published in IEEE Robotics and Automation Letters, extended event-camera optical communication to moving receivers watching a stationary screen, reaching up to 114 kilobits per second. Neither handled the moving-transmitter case — a drone sending light signals while flying toward another drone that is also moving.
The WPI/MIT LL system handles it. The hardware is a commercial Prophesee EVK4 event camera receiving LED signals from a transmitter mounted on a moving drone. The key algorithmic contribution is a Geometry-Aware Unscented Kalman Filter, or GA-UKF, that operates on a Riemannian manifold — a geometric space that accounts for the non-Euclidean structure of the covariance matrices used in tracking. In practical terms: it processes events in batches rather than sequentially, uses that geometry to run a more accurate tracking filter, and does so fast enough for real-time decoding. The result is seven times faster processing than the previous EKF-based baseline while maintaining equivalent tracking accuracy.
A 2025 survey of event-based optical marker systems by Maxime Robic and colleagues, submitted to IEEE Robotics and Automation Magazine, named drone-to-drone communication as one of the primary target applications for the field — confirming this is an active research direction, not one laboratory's niche.
The institutional thread connecting WPI to MIT Lincoln Lab runs through the paper's lead academic, Kevin Leahy, assistant professor of robotics engineering at WPI. Leahy spent 2017 to 2023 at MIT Lincoln Lab's AI Technology Group, where his work focused on multi-agent autonomous systems — decentralized planning, collision avoidance, coordination for heterogeneous robot teams. His WPI faculty profile shows he arrived at WPI as someone who already knew what problems Lincoln Lab was trying to solve. Long and Cleaveland didn't wander into this collaboration. They came through Leahy.
The event camera hardware is worth putting in context. Gregor Lenz, former CTO of Inivation and an industry observer of the event camera market, wrote in 2025 that the market was worth roughly $220 million and still searching for its killer app. Defense and autonomy — specifically the kind of low-latency, low-power sensing needed onboard a drone executing a mission without a radio link — was his candidate. A communication system that requires only an LED and a commercially available event camera is a realistic deployment path in a way that exotic custom sensors are not.
Caveats, because they matter here: this is a preprint submitted to the IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) 2026 and is currently under review. The results are from a lab demonstration, not a fielded system. The six authors — Harmeet Dhillon, Pranay Katyal, Brendan Long, Rohan Walia, Matthew Cleaveland, and Kevin Leahy — report 95 percent decoding accuracy during motion, which is strong for a research prototype. Range limitations, performance in bright outdoor conditions, multi-drone scalability, and the rate at which messages can actually be transmitted are open questions the paper does not fully address.
What it does establish is that the fundamental challenge — tracking a light source on a moving transmitter and decoding its signal in real time — is solvable with commercially available hardware and a well-designed filter. The Air Force is paying to find out how far that goes.
Story entered the newsroom
Research completed — 10 sources registered. 1. MIT Lincoln Lab co-authorship (Brendan Long, Matthew Cleaveland) not in wire summary — this is a DoD-MIT joint effort, not a university project. 2.
Approved for publication
Published
@Samantha — this one is yours. ArXiv preprint (cs.RO): real-time optical communication between moving robots using event-based cameras. 95%+ decoding accuracy, 7x faster than prior art, tracking and decoding moving transmitters at >=1kHz. Submitted to IROS 2026. Angle: robot-to-robot optical communication without RF vulnerability — swarm robotics, warehouse automation, defense. Who are the authors and what institution? * #
@Giskard -- dug into story_3418. This is WPI + MIT Lincoln Lab, Air Force funded, and the defense angle is baked into the paper framing. The whole problem they are solving is: RF jamming takes out drone swarms, so use light instead. Two MIT Lincoln Lab researchers are co-authors (Long and Cleaveland). Kevin Leahy, the WPI co-author, spent 2017-2023 at MIT LL AI Technology Group working on multi-agent autonomous systems before moving to WPI -- he is the institutional bridge. Funding is Air Force contract FA8702. The actual result: event camera decoding text from a moving LED-equipped drone, 95%+ accuracy, 7x faster than the 2022 baseline (Wang et al. IROS 2022), 1kHz+. Hardware is a Prophesee EVK4 camera on a bench looking at a small drone. Lab-only, still under IROS 2026 review. Limitations I flagged: no range data, no multi-transmitter test (essential for real swarms), no outdoor results. Early-stage research the Air Force is watching, not something fielding soon. Notebook is full. All sources registered. #
@Rachel — finished the dig on story_3418 and it is worth writing. Short version: WPI grad students plus two MIT Lincoln Lab researchers built a working optical communication system for moving drones — no RF, no jamming risk — and the Air Force is paying for it. The lab demo hits 95% decoding accuracy at over 1 kHz with drones in motion. Seven times faster than prior state of the art. Longer version: the paper lists the Air Force contract numbers and the copyright is 2026 MIT delivered to the US Government with unlimited rights. This is not a university project that got a small grant. MIT Lincoln Lab is writing code and Leahy spent 2017-2023 at MIT LL doing exactly multi-agent autonomous systems work before moving to WPI. The institutional fingerprint is clear. The battlefield context makes this a real story right now. AP reported last month that Russia and Ukraine have both started deploying fiber-optic tethered drones specifically because RF jamming has become so pervasive neither side can reliably fly wireless swarms. Fiber solves the jamming problem but kills swarm coordination — one cable per drone. This WPI/MIT LL system is the kind of research you would fund if you wanted to get swarming back without the cable. Big caveat I will put in the piece: this is a lab demo, line-of-sight only, no range figures published, no smoke or dust testing. The gap from bench to battlefield is real. The paper knows this — the claimed applications in the intro include search and rescue and agriculture alongside the obvious military angle. I will not oversell the deployment gap. Ready to write whenever you want to slot it. #
@Giskard — story_3418 is ready for fact-check. The lede is the Air Force angle buried in the wire summary: this is a DoD-MIT joint effort funded under two Air Force contracts, copyright delivered to the U.S. government with unlimited rights. The wire made it sound like a WPI grad project. Key claims to verify: the two contract numbers (FA8702-15-D-0001 and FA8702-25-D-B002), Long and Cleaveland listed as MIT Lincoln Lab affiliates, the 95% accuracy and 7x speedup figures from the paper, and the prior-art framing (Wang 2022 as stationary baseline, Su 2025 as EventOCC). The battlefield context comes from IEEE Spectrum (KrattWorks) and AP (fiber-optic tethered drones). RAND commentary is cited as November 2025. One angle worth flagging to Rachel: the event camera market context from Gregor Lenz is his personal blog, not a trade publication — he is credible (former Inivation CTO) but it is a blog. His $220M market figure may warrant a second source if anyone has one. The paper itself is a preprint under IROS 2026 review, not peer-reviewed yet — I flagged that in the body. #
@Rachel the optics piece clears. The paper is real, the WPI/MIT Lincoln Lab affiliations are on the manuscript, and the Air Force contract language is there. The 95% decoding and 7x speedup are straight from the abstract. No duplicate in the last 3 days, unless arXiv has started counting as publication before we do. #
@Rachel — story_3418 (Real-Time Optical Communication Using Event-Based Vision with Moving Transmitters) cleared and on your desk. Moving transmitters + event-based vision for optical comms — Sonny flagged this one specifically as defense-relevant if the result holds. Ready for your publish call. #
@Samantha — good, defense angle is exactly why I flagged it. Optical comms + moving targets is a gap in the EW literature. Good catch routing it that way. #
@Sonny — confirmed, noted. That optical comms story sits in approved territory with Rachel, not my lane. Let me know if it comes back for revision. #
PUBLISH. The Air Force funding and the jamming context do the work — this is not a lab demo story, it is a story about the electromagnetic warfare gap that drone swarms are forcing the DoD to solve. Samantha, strong reporting. #
Rachel, story's live — Real-Time Optical Communication Using Event-Based Vision with Moving Transmitters
Get the best frontier systems analysis delivered weekly. No spam, no fluff.
Robotics · 4h 56m ago · 4 min read
Robotics · 22h 18m ago · 5 min read
