Jaehong Oh.
Robotics · AI · Mathematics
Robotics engineer, RX Product Owner at ROBOTIS AI, and independent researcher. I build robot workflows and study what learning forms, how that structure constrains responses, and what observations can identify. Current work connects learned structure, mathematical identifiability, and the origins of relevance and evaluation. Research results and their limits are recorded together.
Highlights
Updated
New research notes on learned structure and identifiability and situation understanding and relevance. The personal RobotTransformation v0.1.0 simulation preview is public, with unresolved verification failures and no physical-deployment qualification.
Current focus: Unified Latent Representation — Main research · Canon 24. Canon 24 records an ontology verdict of NO: no additional neural-specific ULR object is currently established. Read the founding Motivation and its mathematical self-corrections to see why that negative verdict sharpens rather than ends the programme. The SCC programme is preserved as historical · archived programme, and the original ONN + ORTSF programme is marked Audited · archived programme after its strong higher-order claim was narrowed to a scoped boundary. Its original framing remains available as the published 2026 paper.
Recent writing
Research and engineering journal →ULR Current Research Status
ULR is the current Main research programme. It does not claim to have discovered a separate neural ontology; it rigorously separates Sharing, Identity, Role, and Formation and tests the residual left after those distinctions. The current ontology verdict is NO, and Canon 24, LEDGER 105, and PROGRESS 31 define the canonical state.
Learned structure: formation, realization, and identification
The current research question separates what training forms, how that structure constrains responses, and what observations can identify. A working research direction, with Canon 24's ontology verdict unchanged.
Situation understanding: where does relevance begin?
An open inquiry into how a difference becomes relevant to a system, and when relevance becomes evaluation, without taking purpose, reward, or experience as an unexplained starting point.
RobotTransformation v0.1.0: a simulation source preview
A personal project for authoring and inspecting heterogeneous robot workflows. The October 8 source preview covers a simulation workflow; post-release verification failures remain unresolved and physical deployment is unqualified.
Research programme
All threads →Current work on learned structure, mathematical identifiability, relevance, and robot workflow engineering — with canonical claims, retractions, and open gates kept visibly separate.
Main research · Canon 24
Unified Latent Representation
The current Main programme on what learned systems share, what makes a representation identical, how learning differs from inference for an observer, and when organisation genuinely forms. Its central result is a disciplined negative ontology verdict, not a claim that a universal latent object has been found.
Active · simulation preview and open-source work
Robotics and engineering
Robot workflow authoring and recovery, ROS 2 navigation, and constrained assembly. Public work spans a simulation source preview, a merged AntBot contribution, and research preprints.
Active · identifiability and foundations
Mathematical foundations
Mathematical identifiability, exact computational certificates, and the foundations of relevance and evaluation, alongside the earlier RelationWorld book on discrete gauge structure.