UK defence · autonomous systems · August 2026
Latency-Bounded Perception
Public summary of an analysis of reaction-time limits in thermal perception for autonomous systems. The work was reviewed and not taken forward.
The Problem
Sensors are specified by what they can see: how far, how finely, how reliably they tell one thing from another. Control loops are limited by something else entirely — when the sensor can tell you. A machine that must react cannot act on what the world is doing now, only on the newest picture it has finished forming.
That delay is a property of the detector's physics rather than of how well it is built, and it is written into almost no requirement, test plan, or acceptance criterion.
The Analytical Move
A published requirement asks a system to hold a manoeuvre within three metres of a target closing at roughly the speed of sound. It is read almost universally as a detection and discrimination problem. Dividing the two numbers the requirement already states turns it into a timing problem: three metres is worth under nine milliseconds.
Set the sensor's published physics against that budget and a single stage of the chain — the detector settling on its own, before one picture has been completed — can consume the whole allowance. No improvement in sensitivity, optics, or detection software touches it.
What Is Withheld
The work did not stop at the diagnosis. It proposed a route to recover usable reaction rate beyond the apparent limit, and a three-month bench experiment that would demonstrate that recovery or kill the idea.
The mechanism is not published here. The concept was not taken forward, so it remains unpublished and wholly owned. What is shown is the analysis and the evidence discipline; the method is available under discussion.
What The Pack Demonstrated
- A requirement reframed from its own published numbers, not from assumption.
- An end-to-end timing budget with estimated stages labelled as estimates.
- An experiment scored against a physical reference rather than appearance.
- Three named results, written in advance, that would disprove the concept.
- Readiness and single-practitioner limits stated rather than implied.
Why It Belongs Here
No detailed reasoning was given for the decision, so none is inferred. What survives it is the work: the reframing, the evidence discipline, the declared limits, and the judgement about which part to publish and which part to keep.
The work was produced independently, in response to a requirement the organisation published openly, and offered to it for consideration. It was not a commissioned engagement.
The Documents
Public case note · 7 pages
The reasoning, in full
The reframing and the arithmetic behind it, the experiment proposed to prove or kill the idea, the limits declared at the time, and the outcome.
Open the case noteSubmitted document · not published
Withheld, deliberately
The document sets out the recovery method itself. It was not taken forward, so the method is unpublished and wholly owned, and publishing it as a portfolio piece would place it in the public domain for no return. It is available under discussion.
Where a document can be released in full it is published here in full — the Community Underwater Ordnance Mapper review is an example. This one cannot be, and saying so is more useful to a reader than a document with its argument cut out of it. Nothing shared in confidence is reproduced, and intellectual property in the concept is retained in full.