RuField MFS: An Open-Source Specification for Multimodal Perception in Camera-Free Scenarios

RuField MFS is an open-source specification for multimodal field perception in camera-free scenarios, supporting the fusion of multiple modalities such as WiFi, radar, ultrasonic, infrared, and quantum sensors. Its core features include privacy protection (six-level classification), traceable data provenance (SHA-256 + Ed25519), and implementation in Rust to ensure performance and security. The project is maintained by ruvnet and hosted on GitHub (link: https://github.com/ruvnet/rufield), released on June 14, 2026. It aims to address the privacy issues of traditional cameras and the pain point of lacking a unified abstraction layer for non-visual perception technologies, providing a unified event model to support cross-modal fusion.

Background

Environmental perception is crucial in scenarios like smart homes and smart elderly care, but traditional camera solutions face privacy challenges (user concerns, data risks, regulatory compliance). Although non-visual perception technologies (WiFi CSI, millimeter-wave radar, etc.) can serve as alternatives, they have developed independently, lacking a unified abstraction layer and data model. This leads to difficulties in cross-modal fusion, weak privacy governance, and complex system integration.

Purpose of the Project

RuField MFS proposes an open specification to provide a unified, privacy-aware, provenance-traceable, and fusion-supporting event model for multimodal perception in camera-free scenarios.

Layered Architecture

Adopts a layered design, abstracting heterogeneous perception modalities into standardized field events to enable cross-modal fusion and privacy governance.

Unified Multimodal Abstraction

Supports 15 modalities (WiFi CSI, UWB, millimeter-wave radar, etc.), converting them into standardized field events through core structures like FieldEvent and FieldTensor. Upper-layer applications do not need to focus on hardware details.

Privacy Classification Mechanism

Defines six levels of privacy classification: P0 (raw data is prohibited from leaving the edge by default) to P5 (identity-bound information). The privacy guard automatically decides transmission based on the level, destination, and authorization.

Provenance Traceability and Verifiability

Implements provenance traceability using SHA-256 hashing and Ed25519 signatures; each field event carries a verifiable receipt. Only valid receipts or synthetic data can enter the fusion engine, ensuring input quality.

Implemented in Rust, leveraging memory safety and concurrency features to build a high-performance pipeline, organized via Cargo workspaces:

• rufield-core: Defines core abstractions (Modality enum, FieldTensor, etc.), shared by all components.
• rufield-adapters: Sensor adapters, including synthetic simulators and CsiReplayAdapter (replays real WiFi CSI data).
• rufield-fusion: Fusion algorithm based on weighted Bayesian inference and time windows, outputting high-level scene inferences.
• rufield-privacy/provenance: Implements privacy policies and provenance traceability functions respectively.

Provides a camera-free smart room demo scenario, simulating human activity sequences (entry, sitting down, breathing, falling asleep, etc.), and real-time state inference through collaborative perception of WiFi CSI, millimeter-wave radar, and thermal infrared. The demo can be visualized via a web dashboard (real-time status, event streams, fusion graphs, provenance receipts), and has important application value in scenarios like smart elderly care, medical monitoring, and intrusion detection.

1. Open Specification Design: Architecture decisions are documented via ADR documents, laying the foundation for community participation and standardization.
2. Privacy-First Philosophy: Privacy protection is embedded in the data model layer to achieve fine-grained control.
3. Verifiability Design: Cryptographic methods ensure trusted data provenance, supporting applications in sensitive scenarios.
4. Rust Language Advantages: Ensures system performance and reliability, suitable for edge device deployment.

Current Limitations

The v0.1 version is mainly validated using synthetic data. Although it supports real WiFi CSI replay, full real-time hardware adaptation and labeled dataset validation are pending development. The web dashboard only provides read-only demos, and device management functions are not implemented.

Future Directions

Expand more hardware adapters, improve fusion algorithms, enhance the flexibility of privacy policies, and promote community standardization of the specification.

RuField MFS provides a forward-looking technical framework for multimodal perception in camera-free scenarios, addressing key challenges in heterogeneous modal fusion through unified abstraction, privacy classification, and provenance traceability. Although in the early stage, its open specification concept and solid engineering implementation make it a noteworthy open-source project in this field.