Zero-Onboard-AI-Compute Embodied Intelligence: A General Framework for Collective Computing Through Idle Resources of Other Embodied Agents
Abstract
Embodied intelligent systems are commonly designed with onboard computing hardware capable of executing their own artificial intelligence workloads. This design creates a potential source of hardware redundancy: some agents may possess substantial computing capacity that remains underutilized, while other agents may require intelligent capabilities but cannot justify the cost, power consumption, or physical space associated with a complete onboard AI computing platform. This paper proposes Zero-Onboard-AI-Compute Embodied Intelligence (ZAC-EI), a general framework in which a subset of embodied agents does not carry a general-purpose onboard AI computing platform and instead obtains its intelligent computation from other embodied agents through a shared distributed computing network.
The proposed framework distinguishes computing providers from computing-dependent agents. Providers execute their own mandatory workloads and dynamically contribute surplus processing capacity to the collective network. Computing-dependent agents supply sensor observations, task descriptions, and relevant environmental information to eligible providers, receive computed results, and translate those results into physical actions through their local actuation interfaces. The framework aims to shift AI computation from an individually provisioned device-level resource to a collectively provisioned system-level resource.
Unlike conventional task-offloading architectures that assume every participating device retains its own general-purpose computing capability, ZAC-EI explicitly investigates system configurations in which some agents have no onboard general-purpose AI processor. A formal model is developed for resource allocation, communication delay, provider selection, task deadlines, reliability, hardware provisioning, and network failure. A distributed protocol and a resource-allocation algorithm are proposed. The paper also defines an experimental methodology for comparing the proposed architecture with conventional onboard computing, edge computing, and vehicle-assisted computing.
The framework does not assume that the idea of sharing idle vehicle computation is new. Existing vehicular edge computing research has already explored idle vehicle resources and cooperative task offloading. Its proposed contribution is the explicit investigation of an embodied-agent population containing devices without general-purpose onboard AI computing platforms, together with the architectural, optimization, and reliability requirements associated with that design. The effectiveness and novelty of this approach remain to be established through systematic literature comparison and experimental validation.
Related
- ZAC-Memory: Collective Provisioning of Memory and World Models for Embodied Intelligence
- ZAC-Perception: Collective Provisioning of Perceptual Capabilities in Multi-Agent Embodied Intelligence
- ZAC-Energy: Collective Energy–Computation Coordination for Resource-Constrained Embodied Intelligence
- Beyond Individual Ownership: ZAC, a General Framework for Collective Provisioning of Resources and Capabilities
- Zero-Local-Provisioning (ZLP): A General Framework for Collective Provisioning of Resources, Capabilities, and Attributes
BibTeX
@misc{zhang2026zeroonboardaicompute,
title = {{Zero-Onboard-AI-Compute Embodied Intelligence: A General Framework for Collective Computing Through Idle Resources of Other Embodied Agents}},
author = {Jincheng Zhang},
howpublished = {Zenodo},
year = {2026},
month = oct,
doi = {10.5281/zenodo.23448206},
url = {https://zenodo.org/records/23448206},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Zero-Onboard-AI-Compute Embodied Intelligence: A General Framework for Collective Computing Through Idle Resources of Other Embodied Agents
AU - Jincheng Zhang
PY - 2026
DA - 2026/10/18
PB - Zenodo
DO - 10.5281/zenodo.23448206
UR - https://zenodo.org/records/23448206
AB - Embodied intelligent systems are commonly designed with onboard computing hardware capable of executing their own artificial intelligence workloads. This design creates a potential source of hardware redundancy: some agents may possess substantial computing capacity that remains underutilized, while other agents may require intelligent capabilities but cannot justify the cost, power consumption, or physical space associated with a complete onboard AI computing platform. This paper proposes Zero-Onboard-AI-Compute Embodied Intelligence (ZAC-EI), a general framework in which a subset of embodied agents does not carry a general-purpose onboard AI computing platform and instead obtains its intelligent computation from other embodied agents through a shared distributed computing network. The proposed framework distinguishes computing providers from computing-dependent agents. Providers execute their own mandatory workloads and dynamically contribute surplus processing capacity to the collective network. Computing-dependent agents supply sensor observations, task descriptions, and relevant environmental information to eligible providers, receive computed results, and translate those results into physical actions through their local actuation interfaces. The framework aims to shift AI computation from an individually provisioned device-level resource to a collectively provisioned system-level resource. Unlike conventional task-offloading architectures that assume every participating device retains its own general-purpose computing capability, ZAC-EI explicitly investigates system configurations in which some agents have no onboard general-purpose AI processor. A formal model is developed for resource allocation, communication delay, provider selection, task deadlines, reliability, hardware provisioning, and network failure. A distributed protocol and a resource-allocation algorithm are proposed. The paper also defines an experimental methodology for comparing the proposed architecture with conventional onboard computing, edge computing, and vehicle-assisted computing. The framework does not assume that the idea of sharing idle vehicle computation is new. Existing vehicular edge computing research has already explored idle vehicle resources and cooperative task offloading. Its proposed contribution is the explicit investigation of an embodied-agent population containing devices without general-purpose onboard AI computing platforms, together with the architectural, optimization, and reliability requirements associated with that design. The effectiveness and novelty of this approach remain to be established through systematic literature comparison and experimental validation.
ER -