Visual intelligence
Detection, segmentation, depth and tracking that survive bad light, motion blur and occlusion — validated on real footage, not benchmark sets.
Frontier research lab — Dubai, UAE
YTK Research works on the harder edge of intelligent systems — machine perception, autonomous flight, radio-frequency engineering and agentic AI. Four divisions take the work from first principles through to hardware, software and market.
Each division keeps its own specialism, vocabulary and way of working. They share the same lab, the same engineering standards and the same commitment: build what can be demonstrated, not what sounds impressive.
Frontier work fails when it stays in a paper, and products fail when they are built on assumptions nobody tested. We move along one continuous path, and every stage has to produce something that can be measured.
Decompose the problem until the real constraint appears — physics, latency, cost, or the human in the loop.
Build the smallest system that can fail meaningfully. Instruments and logs from day one.
Edge cases, thermal limits, network loss, adversarial inputs, evaluation suites, guardrails.
Ship into the field with observability, then keep iterating against real operating data.
Six technical territories, each staffed inside a division but solved with the whole lab. Most real programmes touch at least three of them at once.
Detection, segmentation, depth and tracking that survive bad light, motion blur and occlusion — validated on real footage, not benchmark sets.
Airframes, flight control, mission planning and beyond-line-of-sight operation, including inspection payloads and precision positioning.
RF front-end design, link budgeting, mesh and long-range telemetry, interference analysis and spectrum-aware operation.
Planners, tool use, memory, retrieval and multi-agent orchestration — with evaluation harnesses that catch regressions before users do.
Quantisation, pruning and scheduling so models run inside thermal and power budgets on the device, without a round trip to the cloud.
Operator control, teleoperation, trust and handover between autonomy and the person responsible for the outcome.
Whether you need a research partner, a build team or someone to make sense of an architecture that has stopped scaling — the entry point is the same conversation.
Joint work on perception, autonomy and agentic systems — scoped around a concrete technical question rather than a vague innovation mandate.
Hardware and software built by the lab — from edge inference pipelines and drone platforms to production agent deployments.
Architecture review, build-versus-buy, vendor selection and roadmap pressure-testing, delivered by practitioners.
The lab is in Dubai. The work is not limited to it.