Privacy-by-Design Closed-Loop Autonomous Driving: Unifying Execution-Bound Data Erasure, Parameter-Inverted Digital Twins, Safety-Weighted Reinforcement Learning, and Lineage-Bound Unlearning (Paper & 1-Minute Self-Verification Suite)

Privacy-by-Design Closed-Loop Autonomous Driving: A Four-Layer Architecture for Secure Data Erasure, Digital Twins, Reinforcement Learning, and Machine Unlearning Research Overview This preprint presents an integrated four-layer autonomous driving architecture developed by Irum Holdings Co., Ltd., combining privacy-preserving machine learning, verifiable data erasure, parameter-inverted digital twins, safety-weighted reinforcement learning, and model-lineage-based machine unlearning. The proposed framework comprises: Layer 1 — Verifiable Data Erasure: Training-execution-bound data deletion using cryptographic verification and Proof of Secure Erasure (PoSE). Layer 2 — PII-Free Digital Twins: Reconstruction of driving simulation environments using model-derived representations without retaining original personally identifiable driving data. Layer 3 — Safety-Weighted Reinforcement Learning: Closed-loop policy improvement using human corrective feedback, parallel digital twins, and safety-constrained reinforcement learning. Layer 4 — Machine Unlearning and Lineage Verification: Model-level forgetting verification, lineage tracking, and safe rollback mechanisms. The research includes numerical simulations, adversarial verification tests, digital twin evaluations, and machine unlearning experiments. The reported results represent software-level experimental findings rather than validated real-world autonomous vehicle performance. Independent replication, hardware-in-the-loop testing, and full-scale vehicle validation remain necessary. Download the Research Code and Verification Package Google Drive — Complete Research and Verification Package https://drive.google.com/file/d/1mg6WGt3kxW8nekY-hBmJNSrVM6aZ2jb7/view?usp=sharing The downloadable ZIP package contains: Python reference simulations (Phases 1–4) Experimental validation scripts (Exp A, B, and D) Solidity smart contract (PoSEOnChainGate.sol) Automated verification runner (run_all.sh) Research manuscript and LaTeX source README, Dockerfile, and software licensing information Reproducing the Experiments Download and extract the ZIP archive, install the required Python dependencies, and execute: bash run_all.sh The script runs the four reference simulation phases and synthetic verification or smoke tests for Experiments A, B, and D. The Solidity smart contract is supplied separately for inspection. These verification scripts do not constitute full-scale CARLA, nuPlan, hardware TEE, or physical vehicle validation. Intellectual Property and Licensing Patent Holder: Irum Holdings Co., Ltd., Republic of Korea Inventor and Author: Jinwoo Baek The research is associated with Korean granted patent No. 10-3048674 and additional pending Korean patent applications. Three separate intellectual property provisions apply: Research Manuscript: CC BY-NC-ND 4.0 applies to the paper's text and figures. Research Software: Source code is provided under the Irum Holdings Academic & Non-Commercial Verification Software License v1.0, as specified in LICENSE.txt. Patent Rights: All patent rights are expressly reserved. Public research disclosure does not grant any commercial patent license. Commercial deployment, software integration, or patent licensing requires a separate written agreement with Irum Holdings Co., Ltd. Research Collaboration Irum Holdings welcomes independent academic verification, scientific discussion, and research collaboration concerning privacy-preserving autonomous driving, digital twin learning, and machine unlearning. Keywords: Autonomous Driving, Machine Unlearning, Digital Twin, GDPR, EU AI Act, Proof of Secure Erasure, Reinforcement Learning, Privacy-Preserving AI, Model Lineage, Safe OTA.

Authors

Publication Details

Journal
Zenodo (CERN European Organization for Nuclear Research)
Published
2026-11-18
DOI
https://doi.org/10.5281/zenodo.24336851
Primary Topic
Privacy-Preserving Technologies in Data
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Privacy-by-Design Closed-Loop Autonomous Driving: Unifying Execution-Bound Data Erasure, Parameter-Inverted Digital Twins, Safety-Weighted Reinforcement Learning, and Lineage-Bound Unlearning (Paper & 1-Minute Self-Verification Suite)

Jinwoo Baek
2 links
Zenodo (CERN European Organization for Nuclear Research)
Privacy-Preserving Technologies in Data
preprint

Privacy-by-Design Closed-Loop Autonomous Driving: Unifying Execution-Bound Data Erasure, Parameter-Inverted Digital Twins, Safety-Weighted Reinforcement Learning, and Lineage-Bound Unlearning (Paper & 1-Minute Self-Verification Suite)

Jinwoo Baek
preprint en

Abstract

Privacy-by-Design Closed-Loop Autonomous Driving: A Four-Layer Architecture for Secure Data Erasure, Digital Twins, Reinforcement Learning, and Machine Unlearning Research Overview This preprint presents an integrated four-layer autonomous driving architecture developed by Irum Holdings Co., Ltd., combining privacy-preserving machine learning, verifiable data erasure, parameter-inverted digital twins, safety-weighted reinforcement learning, and model-lineage-based machine unlearning. The proposed framework comprises: Layer 1 — Verifiable Data Erasure: Training-execution-bound data deletion using cryptographic verification and Proof of Secure Erasure (PoSE). Layer 2 — PII-Free Digital Twins: Reconstruction of driving simulation environments using model-derived representations without retaining original personally identifiable driving data. Layer 3 — Safety-Weighted Reinforcement Learning: Closed-loop policy improvement using human corrective feedback, parallel digital twins, and safety-constrained reinforcement learning. Layer 4 — Machine Unlearning and Lineage Verification: Model-level forgetting verification, lineage tracking, and safe rollback mechanisms. The research includes numerical simulations, adversarial verification tests, digital twin evaluations, and machine unlearning experiments. The reported results represent software-level experimental findings rather than validated real-world autonomous vehicle performance. Independent replication, hardware-in-the-loop testing, and full-scale vehicle validation remain necessary. Download the Research Code and Verification Package Google Drive — Complete Research and Verification Package https://drive.google.com/file/d/1mg6WGt3kxW8nekY-hBmJNSrVM6aZ2jb7/view?usp=sharing The downloadable ZIP package contains: Python reference simulations (Phases 1–4) Experimental validation scripts (Exp A, B, and D) Solidity smart contract (PoSEOnChainGate.sol) Automated verification runner (run_all.sh) Research manuscript and LaTeX source README, Dockerfile, and software licensing information Reproducing the Experiments Download and extract the ZIP archive, install the required Python dependencies, and execute: bash run_all.sh The script runs the four reference simulation phases and synthetic verification or smoke tests for Experiments A, B, and D. The Solidity smart contract is supplied separately for inspection. These verification scripts do not constitute full-scale CARLA, nuPlan, hardware TEE, or physical vehicle validation. Intellectual Property and Licensing Patent Holder: Irum Holdings Co., Ltd., Republic of Korea Inventor and Author: Jinwoo Baek The research is associated with Korean granted patent No. 10-3048674 and additional pending Korean patent applications. Three separate intellectual property provisions apply: Research Manuscript: CC BY-NC-ND 4.0 applies to the paper's text and figures. Research Software: Source code is provided under the Irum Holdings Academic & Non-Commercial Verification Software License v1.0, as specified in LICENSE.txt. Patent Rights: All patent rights are expressly reserved. Public research disclosure does not grant any commercial patent license. Commercial deployment, software integration, or patent licensing requires a separate written agreement with Irum Holdings Co., Ltd. Research Collaboration Irum Holdings welcomes independent academic verification, scientific discussion, and research collaboration concerning privacy-preserving autonomous driving, digital twin learning, and machine unlearning. Keywords: Autonomous Driving, Machine Unlearning, Digital Twin, GDPR, EU AI Act, Proof of Secure Erasure, Reinforcement Learning, Privacy-Preserving AI, Model Lineage, Safe OTA.

Zenodo (CERN European Organization for Nuclear Research)
Industry, innovation and infrastructure
Privacy-Preserving Technologies in Data
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon