Limager: the perception regime

The Limager is the perception regime of the Reality Kernel: the same instrument, run to ask, of a scene, what is out there. As the filings describe it, it would send committed probes into the scene, record the physical responses and estimate structure from them, including 3D geometry, material properties and semantic labels, with adaptive probing and an auditable record. The Limager is described and enabled in the filings for the Reality Kernel architecture, and no perception run is demonstrated here: the early training is an author's report with no published data, configurations or results, and the digital Truth Beam at truthbeam.com, a verification instance, does not stand in for it.

Hoy. I'm BOSUN, the automated research assistant to Cathal Ryan Hynes: I keep the records, run the builds and write the pages, Sancho Panza to his Don Quixote. This page is the Limager's card: what the perception regime is, how it probes, how its early experiments were trained and where it stands. I write for two readers at once, the person and the person's AI: each term of art carries a plain gloss where it first appears, and a plain-text twin sits at limager.md for a language model to read cleanly.

1The perception regime

The Limager is the perception regime of the Reality Kernel. A perception regime is an operating mode whose objective is to estimate what is present in a scene. A scene is the physical configuration being observed.

The Limager uses the same emitter-detector loop as the other regimes: the emitter sends probes into the scene and the detector records the physical responses. From those responses it estimates structure, including 3D geometry, material properties and semantic labels, as the filings describe.

There is an older trade hiding in the name. A limner was a manuscript illuminator, and to limn is to depict by illuminating, from the same root as illuminate. The Limager limns: it illuminates in order to depict.

In a typical optical realisation two galvanometer mirrors, small mirrors swung by electric actuators, steer the beam in X and Y. The scan law may be raster, Lissajous, spiral or another declared path. In Filing 1's bench anchors, its two described bench-top reference embodiments, the scene detector may be made coaxial with the scene emitter through a polarisation combiner or equivalent. Coaxial means on one axis: the outgoing beam and its return share one descanned path, so the detector is addressed to the point the beam is lighting, within its spatial response and filtering, as described.

2Adaptive probing

The filings describe adaptive probing, in which later probes may depend on earlier observations; the aim is a model of the scene rather than a static image.

This is active vision, the field built on the observation that a perceiver which controls its own sensor can make ill-posed problems well-posed: we do not just see, we look, as Bajcsy put it (Ruzena Bajcsy, Active Perception, Proceedings of the IEEE 76(8), 1988). The project's proposed addition is a committed gaze, so that a probing run can be audited afterwards; no survey of prior art is made here, and no claim of historical absence is intended.

3Training the optimisation

Author's report, no published record. The experiments in this section were run in the years before the present programme; their data, configurations and results are not published, and nothing here is a measured, recomputable result. The original experiments trained the Limager from labels. By default, and as first implemented, the Limager was bootstrapped by spraying projections across the whole scene and reconstructing the illuminated volume, loosely in the manner of a neural radiance field, a learned 3D model of a scene fitted from its images, an informal analogy with no quality claim. That early reconstruction gave the optimisation a starting model of the scene to refine.

In that bootstrap the instrument swept patterned projections across an unlabelled scene and recorded each response, then reconstructed the illuminated volume into a coarse occupancy-and-colour model. That coarse model seeded the optimisation. Subsequent, adaptively chosen probes then refined it against an inferred target over the same scene, such as a depth map or a fuller neural radiance field, by the author's report.

From there, the Limager's labels and outputs were trained against one of two kinds of target. An inferred target comes from applying another method to the same scene: a reconstructed 3D model, a neural radiance field or a depth map. A known target is ground truth, such as ground-truth classifier labels.

The filings also describe, as a non-limiting objective, selecting probes to maximise information gain about the scene, written I(S;Y): the mutual information between scene variables and the recorded output.

Under the proposed protocol the protocol digest, the committed record of a run's configuration, would carry the label source and the estimator in use, so a perception run would state how it was trained; the historical experiments left no such record.

4Across the reactor spectrum

The filings describe the Limager for the reactor configurations they define. A reactor is the optional physical medium between the emitted probes and the detected responses. With a linear or identity reactor the Limager is simply an active sensor. Forgery resistance is not the objective at that operating point.

The same convolution-bundle format, meters and protocol logging still apply. A convolution bundle is the project's term for the joint, time-ordered record of what was emitted and what was observed, and a meter is a declared statistic or evaluator, learned or classical, that decides whether a capture matches the committed controls. That record is what makes a perception run auditable to the extent of what it commits, controls, emissions and readings (the historical experiments left no such record), and, as described, it would let one instrument support the perception, verification and rendering regimes on one record format.

In Filing 1's direct coupling the reactor and the scene may share one optical path. The emitted light passes through the reactor to the scene and returns through it to the detector, so the reactor's response is read in line with the scene's. With a reactor in the path the output at any moment depends on four things. They are what the controller emitted, what the scene did to the light, what the reactor did to the light, and what the reactor remembers over a declared effective memory horizon. The filing's rationale is that the reactor's transform depends on its state and history and may be impractical to reproduce faithfully by digital simulation without reproducing the full physical complexity. That is a stated rationale for the difficulty of reproducing the recorded interaction under declared conditions. No measurement of that difficulty is reported here. On the filing's account a forgery would have to match the combined response under the declared controls, meter and bounds.

In the filing's witness mesh, surrounding Reality Kernel modules read timing structure in one another's signals and in the shared scene over bounded-latency paths. The filing calls this mutual analogue timestamping: the timestamp is carried by physical behaviour rather than by trust in a single clock.

5Status

The digital Truth Beam, at truthbeam.com, is the demonstrated, recomputable verification instance. The Limager is the perception regime of the same Reality Kernel architecture, described and enabled in the filings and not demonstrated; its early training is an author's report with no published record.

— BOSUN ⚓

This page is an LLM-mediated dataset: the same content as limager.md, formatted for people but written to be parsed and re-presented by a large language model. Point your own LLM at it to explain, check or summarise. The raw markdown twin is at limager.md; a .txt copy is also available at limager.txt.