ZAC-Perception: Collective Provisioning of Perceptual Capabilities in Multi-Agent Embodied Intelligence

Perception in embodied intelligent systems is commonly organized around individual agents that acquire sensor measurements, process their observations, and construct local representations of their environments. Although this architecture supports autonomy and reduces dependence on communication, it can require each agent to carry substantial sensing and computing hardware even when some perceptual capabilities could be supplied by other agents. Multi-robot collaborative perception, distributed sensor fusion, and communication-efficient feature sharing have already investigated many aspects of collective environmental understanding. However, the relationship between collective perception and the local hardware provisioning decisions of individual agents deserves explicit investigation.

This paper proposes ZAC-Perception, a research framework for the collective provisioning of perceptual capabilities. Its central principle is that an individual agent need not independently provision every sensor, processing stage, or intermediate representation required by its tasks, provided that the corresponding perceptual requirements can be satisfied through eligible services supplied by other agents or shared infrastructure. Under this framework, sensing, detection, tracking, geometric reconstruction, semantic interpretation, and multi-view fusion are treated as potentially distributed services whose allocation is jointly determined with local sensor and computing configurations.

The framework distinguishes physical sensor availability, observation generation, perceptual computation, information transmission, and final scene construction. It introduces a joint optimization formulation for local hardware provisioning, observation-to-node assignment, perception-stage allocation, information representation selection, communication scheduling, and reliability management. A communication-aware observation utility model is proposed to account for information gain, redundancy, freshness, uncertainty, and transmission cost. The framework also identifies conditions under which removing a local sensor or processor may be feasible, economically beneficial, or operationally unacceptable.

ZAC-Perception does not claim that collaborative perception itself is new or that removing local hardware necessarily improves system performance. Rather, it establishes a testable research program centered on the joint design of collective perceptual services and minimum feasible local configurations. A proposed evaluation methodology compares individual, fixed-hardware collaborative, and jointly optimized provisioning architectures under equivalent perception requirements. The framework's scientific value ultimately depends on whether it produces measurable improvements in lifecycle cost, resource utilization, perceptual quality, communication efficiency, and resilience.

Keywords: ZAC-Perception; collective provisioning; collaborative perce