Building the spine that will enable autonomy at scale.
RABARA is a research-led venture developing the reasoning architecture that will define how autonomous systems think, decide and act. Grounded in doctoral research. Validated in the field. Built for the long arc.
RABARA operates commercial UAVs in precision agriculture. We have been operational since 2025, flying missions across farms in Kenya — spraying pesticides and spreading fertiliser, with every mission logged as structured operational data.
Our operations are currently conducted under Visual Line of Sight (VLOS) regulations — one pilot, one drone, maintained within direct visual range at all times. It is the industry ceiling. Understanding that ceiling, and building the rigorous case to move beyond it, is what drives everything else we do.
Concurrent with our flight operations, we are building digital twins of the farms we serve. These virtual models — continuously updated from operational data — form our first intelligence layer: a predictive environment for counterfactual reasoning about when to spray, how much to apply, where to place inputs and at what frequency. By connecting specific interventions to measurable outcomes through causal inference, we begin to answer the question every farmer is really asking: what is the best action to take, and why?
UAV-based pesticide application across farmland. Precise, repeatable and fully logged — every run generating structured data on environmental conditions, equipment performance and mission parameters.
Variable-rate spreading guided by field data and agronomic requirements. Efficient delivery for farmers and granular input data feeding our farm digital twin models.
Every operational mission enriches a virtual model of the farm. These digital twins are the foundation of our intelligence layer — enabling causal reasoning about interventions and their outcomes before they are executed in the field.
The VLOS ceiling is not a RABARA problem. It is the constraint facing every serious drone operator. To scale UAV operations — in agriculture, in infrastructure monitoring, or across any complex environment — you need Beyond Visual Line of Sight (BVLOS) capability: autonomous, reliable, routine operations without a pilot maintaining constant visual contact.
What makes this genuinely hard is that risk is not static. It shifts continuously as environment changes, equipment degrades, crew conditions evolve and mission parameters update. A system that cannot track these deltas in real time cannot manage risk — and a system that cannot manage dynamic risk will not earn regulatory acceptance for BVLOS operations. That is the burden of proof. It has not yet been formally met by the industry.
One pilot. One drone. Direct visual contact maintained at all times. The regulatory default that caps every operator's scale, efficiency and commercial reach — regardless of technical capability.
Routine BVLOS operations would transform precision agriculture and infrastructure monitoring alike. The technology broadly exists. The missing piece is demonstrable, certifiable risk awareness — evidence that the system can manage dynamic risk autonomously.
Regulators require evidence — not that a system is capable, but that it is safe. That it can perceive, reason about and manage risk that is itself constantly changing across environment, equipment, crew and mission.
RABARA is developing a hybrid reasoning framework — grounded in active doctoral research — that enables autonomous systems to reason about risk the way expert human operators do: drawing simultaneously on rules, uncertainty, prediction and causal understanding.
The framework extends naturally from the causal reasoning we are already building through our farm digital twins. The doctoral research takes this foundation further — introducing symbolic, probabilistic and predictive reasoning alongside causal inference — producing a richer, more robust architecture capable of managing dynamic risk in high-stakes environments. This is the spine. Built to answer one question: how does an autonomous system earn the right to act?
"Risk awareness is not a feature of intelligent systems. It is the evidence of intelligence itself. The only path to competent autonomy is through reasoning — and the most rigorous object of that reasoning is risk."
— RABARA Research FrameworkRule-based logical inference. Explicit, auditable decision logic applied to known constraints, regulatory requirements and mission parameters — the foundation of certifiable, explainable behaviour.
Formal uncertainty quantification. The system knows what it does not know — weighing every decision against risk thresholds accordingly. Confidence is always declared, never assumed.
Anticipating future states before they occur. The system models evolving conditions — environment, equipment, crew, mission — forward in time, acting on what is coming rather than only what is present.
Understanding why things happen, not just that they do. Causal models allow the system to reason about interventions and their downstream consequences — the basis of genuine judgement and the intellectual bridge from our agricultural work to autonomous systems broadly.
The reasoning architecture operates within a digital twin environment — a continuously updated virtual model of the operational context, whether a farm, a stretch of infrastructure or a broader mission area. Before any consequential decision is executed in the physical world, it is tested in the digital one. This is not a product feature. It is the mechanism by which the system validates its own reasoning — and the means by which we generate the rigorous evidence that regulators will one day require.
Every operational hour is evidence accumulating toward a single goal — a rigorous, validated, data-backed case for autonomous BVLOS operations placed before a regulator. Our two proving grounds run almost simultaneously. Agriculture first, infrastructure monitoring close behind. Both are vehicles for the same mission.
Our operational foundation. Agriculture provides accessible, structured environments to accumulate flight hours, refine data collection and deploy the first versions of our reasoning framework — with real farms, real clients and measurable outcomes.
The use case at the centre of our ongoing doctoral research. Monitoring critical national infrastructure demands exactly the risk-aware, BVLOS-capable autonomy we are building toward — in the environment where the stakes of getting it wrong are highest.
In a few years, we intend to place a validated, evidence-backed case before a regulator — demonstrating that this system understands and manages dynamic risk well enough to fly autonomously, beyond visual line of sight, at scale.
A hybrid reasoning framework proven in autonomous UAVs does not stay in UAVs. The architecture we are building — validated first in agriculture, then in infrastructure monitoring — is designed from the outset for generalisation. Once it demonstrates that it can reason about dynamic risk in complex, high-stakes environments, it becomes the backbone of intelligent autonomous systems far beyond aviation.
The local generalisation is across UAV-based use cases. The global generalisation is to entirely new fields. The question becomes not where else this applies — but where it does not.
At the furthest horizon: precision medicine. The same predictive environment that tests a drone's decisions before real-world execution can simulate proposed medical interventions on a digital twin of the human body — before they are tried in real life. A rigorous, simulation-first approach to drug development and treatment design, where outcomes can be tested safely before they are trusted completely.
Every flight we make is a sentence in an argument we are building for the future of autonomous systems. We are looking for people who want to be part of that argument — as partners, funders and collaborators who understand that the most important question in autonomy is not can it fly, but does it deserve to.