The research behind the runtime
AEROS is built around one idea: an embodied agent that keeps learning needs an identity layer — a frozen, cryptographically verifiable core that fixes who the agent is and what it is permitted to do, decoupled from the competence that keeps growing. Sandbox runtimes isolate what an agent process can touch; the identity layer governs which embodied individual is acting, under whose authority, within which behavioural envelope, with what auditable history. OpenShell governs the infrastructure; AEROS governs the individual.
Governable individuals
Governable Individuals: An Identity Layer for Embodied Agents That Keep Learning
The position piece: why learning agents need a verifiable identity layer, and what it takes to build one.
arXiv:2607.05463 Technical reportGoverned Individuation: Cryptographically Decoupling an Agent's Learning from Its Authority
The companion measurement study: identity invariance, permission conservation, auditability, and convergence, established over a real governance runtime.
arXiv:2607.04613 Codes20sc/governed-individuation
The experiment harness behind the technical report — self-contained, with one-command reproduction and every released result checked in.
github.com →One program, four pillars
The runtime's core mechanisms were each developed and evaluated as a research contribution first, then engineered into the subsystem that ships in the repository. The papers below are grouped by the pillar they built; more will be added as they land on arXiv.
Runtime & Governance
The operating architecture and the admission → policy → contract → safety gate chain — and the benchmark that measures it.
- arXiv:2604.07039 AEROS: A Single-Agent Operating Architecture with Embodied Capability Modules →
- arXiv:2604.07833 Harnessing Embodied Agents: Runtime Governance for Policy-Constrained Execution →
- arXiv:2604.11174 EmbodiedGovBench: A Benchmark for Governance, Recovery, and Upgrade Safety in Embodied Agent Systems →
Capability Contracts & Evolution
Versioned, signed capability interfaces and the validator → shadow → canary upgrade path behind the ECM system.
- arXiv:2604.08059 Governed Capability Evolution: Lifecycle-Time Compatibility Checking and Rollback for AI-Component-Based Systems, with Embodied Agents as Case Study →
- arXiv:2604.13097 ECM Contracts: Contract-Aware, Versioned, and Governable Capability Interfaces for Embodied Agents →
- arXiv:2604.26689 Regression Test Selection for Updated Capability Modules in Compositional ML Systems via Atomic-Quality Probes →
Memory & Identity
Learning that sediments into memory and persona without ever moving the identity hash.
Fleet & Deployment
Multi-robot coordination, identity-stable canary rollout, and governed role reassignment across heterogeneous fleets.
- arXiv:2604.11028 Federated Single-Agent Robotics: Multi-Robot Coordination Without Intra-Robot Multi-Agent Fragmentation →
- arXiv:2605.28097 ICAN-Deploy: Identity-Stable Canary Deployment for Safety-Critical Embodied Agents →
- arXiv:2607.04634 Governed Caste Reassignment in Heterogeneous Swarms: An Asymmetric-Trust Protocol with Audited Operator Countersignature →
Citing AEROS
If you use AEROS in research, please cite the relevant paper(s) above — the papers are the theoretical foundations, the repository is the engineering implementation. For the identity-layer idea as a whole, cite the perspective and the technical report:
@article{qin2026governable,
title = {Governable Individuals: An Identity Layer for Embodied Agents That Keep Learning},
author = {Qin, Xue and Luan, Simin and Yang, Cong and Li, Zhijun},
journal = {arXiv preprint arXiv:2607.05463},
year = {2026}
}
@article{qin2026governed,
title = {Governed Individuation: Cryptographically Decoupling an Agent's Learning from Its Authority},
author = {Qin, Xue and Luan, Simin and Yang, Cong and Li, Zhijun},
journal = {arXiv preprint arXiv:2607.04613},
year = {2026}
}
A CITATION.cff for citing the codebase itself ships in
the repository.