Abstract
As automated driving systems (ADS) proliferate, the need for remote human intervention to handle "corner cases" has led to a fragmented landscape of terminology. This paper surveys standards, corporate approaches, scientific literature, and legislative frameworks to propose a unified six-level taxonomy for Remote Human Input Systems (RHIS). By mapping human input complexity to existing SAE automation levels, the taxonomy provides a common language for engineers, policymakers, and the public.
Introduction: The "Tower of Babel" in Remote Driving
Automated Driving Systems are rapidly improving, yet they are not yet capable of handling every possible road scenario. When a vehicle encounters a "corner case"—such as a construction site with ambiguous hand signals or an undocumented road closure—it often requires assistance from a human located in a remote command center.
However, the industry currently lacks a standardized vocabulary to describe these interactions. Terms like "teleoperation," "remote assistance," "fleet response," and "supervisory control" are often used interchangeably or with conflicting definitions. This "Tower of Babel" effect complicates the development of safety regulations, hinders technical comparisons, and creates confusion in public communication.
Methodology
To unify these disparate concepts, the authors performed a comprehensive survey across four key domains:
- Standards and Technical Reports: Analyzing the current SAE J3016 definitions and identifying gaps regarding remote support.
- Corporate Approaches: Reviewing how industry leaders like Waymo, Zoox, Nissan, and Uber describe their proprietary remote support systems.
- Scientific Literature: Tracing the evolution of "teleoperation" and "telerobotics" from the late 1980s to the present.
- Legislative Frameworks: Examining how governments in Germany, the UK, and the US (specifically California and Arizona) are codifying remote operation into law.
The Terminology Problem
The survey highlights a significant divide between direct control (e.g., steering via joystick) and indirect assistance (e.g., path selection). Current standards like SAE J3016 do not provide sufficient detail to distinguish between these modes, particularly when a human provides high-level strategic input without taking over the dynamic driving task.
A Unified RHIS Taxonomy
The centerpiece of this research is the proposed RHIS Taxonomy, which categorizes remote input into six distinct levels (0-5) based on the complexity and nature of the human intervention:
- Level 0 (Remote Monitoring): Passive observation and diagnosis of system states. This serves as the essential baseline for all other levels and corresponds to SAE Level 3.
- Levels 1 & 2 (Remote Driving): The human performs the Dynamic Driving Task (DDT) directly. Level 2 includes "Safety Guards" where the vehicle's onboard systems can prevent the remote driver from executing unsafe maneuvers. These levels correspond to SAE Levels 1-2.
- Levels 3, 4, & 5 (Remote Assistance): The human provides assistive input while the ADS maintains control of the DDT.
- Level 3: Perception assistance (e.g., identifying a specific object).
- Level 4: Strategic input (e.g., selecting an alternative route).
- Level 5: Authorization (e.g., approving a maneuver suggested by the car). These levels correspond to SAE Levels 4-5.
Defining Operator Roles
The taxonomy also establishes clear roles for personnel within a remote command center:
- Remote Driver: Operates at Levels 1-2, directly controlling vehicle movement.
- Remote Assistant: Operates at Levels 3-5, providing high-level guidance.
- Remote Observer: Operates at Level 0, monitoring system health.
All these roles are unified under the umbrella term Remote Operator (ROp). This classification is "vehicle agnostic," applying equally to conventional cars with remote backups and dedicated robotaxis without manual controls.
Limitations and Future Work
While the RHIS taxonomy provides a robust framework, several areas require further exploration. The functional delimitation between the ADS and the Remote Operator can remain fuzzy in complex edge cases. Furthermore, the impact of network latency on safety across different RHIS levels is not yet fully integrated. Future research should also focus on the human factors involved, such as the psychological workload and specific training requirements for remote operators across different levels of intervention.
Summary updated