Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
SAE J3131-2022: A Reference Architecture for Automated Driving Systems
The rapid evolution of automated driving systems (ADS) has created a pressing need for clear, consistent language and a shared understanding of how these complex systems are structured. SAE International’s SAE J3131-2022, ‘Definitions for Terms Related to Automated Driving Systems Reference Architecture,’ addresses this need by providing a descriptive, informative, and non-normative framework for Level 4 and Level 5 ADS. This recommended practice builds upon the National Institute of Standards and Technology (NIST) 4D-RCS model and the ALFUS framework to decompose advanced automation tasks into manageable, hierarchical layers. The result is a standard vocabulary and architectural blueprint that helps engineers communicate, design, and validate their systems with greater precision and less ambiguity.
🛠️ Engineering Design Insight: The 4D-RCS framework decomposes complex ADS tasks into hierarchical layers based on time-based planning horizons—from long-range strategic planning to immediate reactive control. This decomposition reduces system complexity and simplifies design, verification, and validation.
Purpose and Scope of SAE J3131
SAE J3131 is a recommended practice, meaning it offers guidance rather than mandatory specifications. Its goal is to capture and standardize the language already in use within the ADS development community, while filling gaps where definitions are ambiguous or missing. The document primarily targets Level 4 and Level 5 automation features (as defined in SAE J3016), though some terms and concepts may apply to lower levels. The guiding principles behind J3131 are to be descriptive and informative, provide functional definitions, and remain consistent with both prior art and current industry practice.
The reference architecture illustrated in the standard—and referenced throughout—shows how sensing, sensor processing, fusion, and perception feed into higher-level planning and vehicle control functions. This hierarchical approach allows designers to divide the driving task into distinct functional layers, each with its own spatial and temporal resolutions, making the overall system more manageable and testable.
Core ADS Terminology: Sensing, Fusion, and Perception
A major contribution of SAE J3131 is the precise definition of fundamental terms that are often used interchangeably but have distinct meanings in the context of ADS. The table below summarizes the key functional components defined in the standard.
| Term | Definition (from SAE J3131) |
|---|---|
| Sensing | Use of a device (sensor) to acquire information about an object or a physical phenomenon. |
| Sensor Processing | Functions that operate on sensor signals to compute characteristics of the operating environment, detect, identify, and classify traffic entities, events, and situations. |
| Fusion | Information processing that manages filtering, correlation, comparison, association, and combination of data from multiple sources to produce high-confidence state estimates. |
| Perception | The ADS’s capability to sense and characterize the entities, events, and situations in its environment. Perception produces outputs at multiple levels of abstraction. |
It is important to note that fusion and perception are distinct functions. Fusion handles the mathematical integration of uncertain data, while perception uses the fused data to derive meaning about the environment. The standard further breaks down perception into five levels (P0 to P4) based on aggregation—from individual points to groups of objects.
| Perception Level | Output Description |
|---|---|
| P0 | Point or pixel: a discrete element with measured physical properties. |
| P1 | Line or list: groupings of points by continuity in position and direction. |
| P2 | Surface or boundary: groupings of lines by continuity in orientation or curvature. |
| P3 | Object: groupings of surfaces forming a rigid body. |
| P4 | Groups of objects: collections of objects based on density, distribution, and interactions. |
⚠️ Common Mistake: Do not confuse perception levels (P0–P4) with the levels of driving automation defined in SAE J3016. They are unrelated concepts. Also, fusion and perception are not synonymous—misusing these terms can lead to design errors and regulatory misunderstandings.
Architectural Insights: Hierarchical Planning and Decomposition
The SAE J3131 reference architecture leverages the 4D-RCS (Real-time Control System) framework, originally developed by NIST. This framework decomposes the driving task into layers based on time-based planning horizons: the highest layer might plan a route over minutes, while lower layers handle path selection over seconds and vehicle control over milliseconds. By clearly separating these concerns, engineers can design, test, and modify each layer independently, greatly reducing system complexity.
Another key insight from the standard is the hierarchical nature of perception. The perception function does not produce a single output; instead, it creates a cascade of increasingly abstract representations—from raw points to object groups. This aligns with how human drivers process information and allows different ADS subsystems to consume data at the appropriate level of granularity.
Frequently Asked Questions
Is SAE J3131 mandatory?
No. J3131 is a recommended practice, not a mandatory specification. It provides a common vocabulary and conceptual framework that developers may choose to adopt. However, its adoption can streamline communication and compliance with related standards.
How does J3131 relate to SAE J3016?
J3016 defines the levels of driving automation (0–5). J3131 focuses on the reference architecture and terminology for the automated driving system itself, primarily for Level 4 and Level 5 features. It references J3016 for the definition of ADS.
What is the role of 4D-RCS in this standard?
4D-RCS is a reference model that provides the basis for the hierarchical decomposition used in J3131. Its time-based planning horizon concept is central to the architecture, enabling effective division of the driving task into layers with different spatial and temporal resolutions.
Can J3131 be used for lower levels of automation?
While primarily aimed at Level 4/5, some terms and concepts may apply to lower levels. However, the architecture assumes no driver intervention, so lower-level systems that rely on a human driver may not fully benefit from the reference architecture.
🔍 SAE J3131 is an essential resource for engineers and researchers working on advanced automated driving systems. By adopting a common language and architectural framework, the industry can accelerate development and improve safety.
