Learning from Disengagements: An Analysis of Safety Driver Interventions during Remote Driving
1 st Ole Hans, 2 nd Jürgen Adamy · 2025
Read Learning from Disengagements: An Analysis of Safety Driver Interventions during Remote Driving by 1 st Ole Hans, 2 nd Jürgen Adamy (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-00246.txt.
Introduction
Ole Hans and Jürgen Adamy analyzed safety driver takeovers during remote driving training in 2025. In remote driving, a trained driver runs a car from afar while a safety driver in the car can step in. Data came from a real commercial car-sharing service.
The team studied more than fourteen thousand kilometers of remote driving by twenty-five trainees. Takeovers fell sharply within the first four hundred kilometers of practice, a clear learning curve. They sorted one hundred eighty-three takeovers into four main scenarios that trigger a safety step-in. The results argue for training that builds basic skills early and targets known hard scenarios, above all in complex city areas.
Catalog record
| CREATOR | 1 st Ole Hans, 2 nd Jürgen Adamy |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-00246
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Learning from Disengagements: An Analysis of Safety Driver Interventions during Remote Driving Thanks: This study was conducted in collaboration with Vay Technology, a company operating a remotely driven commercial car-sharing service. The analysis focuses on data collected from operations in Las Vegas, Nevada, US, while Vay Technology also conducts remote driving operations in Hamburg and Berlin. 1 st Ole Hans, 2 nd Jürgen Adamy Abstract This study investigates disengagements of Remote Driving Systems (RDS) based on interventions by an in-vehicle Safety Drivers (SD) in real-world Operational Design Domains (ODD) with a focus on Remote Driver (RD) performance during their driving training. Based on an analysis of over 14,000 km on remote driving data, the relationship between the driving experience of 25 RD and the frequency of disengagements is systematically investigated. The results show that the number of SD interventions decreases significantly within the first 400 km of driving experience, which illustrates a clear learning curve of the RD. In addition, the most common causes for 183 disengagements analyzed are identified and categorized, whereby four main scenarios for SD interventions were identified and illustrated. The results emphasize the need for experience-based and targeted training programs aimed at developing basic driving skills early on, thereby increasing the safety, controllability and efficiency of RDS, especially in complex urban environment ODDs. I Introduction The remote operation of vehicles is becoming increasingly important, especially in the context of modern mobility concepts and the further development of Automated Driving Systems (ADS). Studies such as the remote driving report published by McKinsey [ 1 ] and progress of regulations for remote driving in Germany [ 2 ] underline the relevance of the remote driving technology for the future of mobility. However, in order to safely introduce this technology to the market, a well-founded examination of central research questions is required, as emphasized in the Federal Highway Research Institute (BASt) research report for teleoperation needs [ 3 ] . One of these central questions concerns the requirements for the education and training of Remote Drivers (RD) . Existing standards and guidelines, such as the BSI Flex 1887 standard [ 4 ] , demand comprehensive training that focuses on familiarity with different vehicle conditions and realistic training environments. Likewise, the importance of targeted training to improve human performance and mitigate the inherent challenges of remote driving is emphasized in related research work by several authors [ 5 , 6 ] . However, before specific requirements for the education and training of RDs can be formulated, it is crucial to identify the underlying problems. Previous research has often relied on simulations [ 7 ] or controlled test environments [ 8 ] , often with untrained participants. These approaches provide valuable insights into human behavior under different conditions, but cannot fully reflect the complexity of real-world driving situations [ 9 ] . The gap between industrial applications and scientific validation is also highlighted by the increasing number of implementations in an industrial context, for example by companies such as Vay Technology or Fernride. While these companies demonstrate the practical potential of remote driving, scientific research into human performance in real-world Operational Design Domains (ODD) remains insufficient. This work addresses the gap in the scientific investigation of Remote Driving Systems (RDS) on public in real-world ODD and analyses disengagements of the RDS by in-vehicle Safety Driver (SD) interventions. These are situations in which the vehicle control has to be taken over by the in-vehicle SD to prevent a potential hazardous situations. The aim is to use this data to draw conclusions about the requirements for training, system safety design and the specification of ODDs in order to create the basis for a safe and effective market introduction of remote driving. Section II provides literature research relevant to the study, and Section III describes the RDS in its parts and the respective ODD as the basis for this study. Section IV describes the data set which is used. In Section V the disengagement classification is defined and the disengagement distribution in addition to the impact of RD experience are evaluated. Based on that Section VI presents the analysis results of the disengagement reasons in detail. The limitations of the results are described in Section VII . Finally, the conclusion and proposed further work are presented in Section VIII . II Related Work This Section presents an overview of research related to remote driving, their limitations, and SD disengagements for remote driving. II-A Remote Driving Remote driving in the automotive industry is distinguished from related concepts such as remote assistance and remote monitoring, forming a nuanced taxonomy of Remote Human Input Systems that varies in complexity [ 10 ] . Remote driving involves the direct control [ 11 ] of a vehicle by a RD, from a remote location, enabling the navigation of a vehicle through complex environments without requiring physical presence in the vehicle. Building on foundational work by Bogdoll et al. [ 10 ] and Amador et al. [ 12 ] , this field is classified by levels of complexity, which aligns with the SAE taxonomy for ADS [ 13 ] . The core of remote driving lies in a stable wireless connection between the Remote Control Station (RCS) and the vehicle, enabling real-time monitoring and control. This connection allows the RD to access the vehicle within ODD remotely, using data transmitted from cameras and further sensors, albeit with latency that impacts real-time the RD’s situational awareness [ 14 , 7 ] . Challenges in RDS operation are both technical and human-centered. Technical challenges such as latency, video quality, and visibility impairments create performance constraints for the RDS itself as well as for the RD, while human factors, such as the lack of haptic feedback, are additional factors that reduce the situational awareness of the RD [ 15 , 16 ] . The RD relies mainly on visual feedback, which poses unique limitations compared to conventional driving. Latency and reduced video quality can further impede the RD’s capacity to maintain consistent situational awareness while operating the vehicle [ 8 , 7 ] . Given these limitations, a well-defined ODD must reflect not only system capabilities but also the human driver’s performance threshold to ensure effective RD functionality [ 17 ] . Addressing these challenges demands both functional adaptations and operational measures, emphasizing the need for both technological enhancements and targeted human performance adaption and training to mitigate the inherent limitations of remote driving [ 5 , 16 ] . II-B Disengagements Related studies have investigated disengagements mainly in the context of ADSs, where the in-vehicle SD intervenes when the ADS cannot perform the driving task autonomously [ 18 ] . These studies often focus on the frequency of disengagements, the ODD factors that contribute to such events, and the underlying technical failures. Kalra and Paddock [ 19 ] analyzed disengagement reports from tests of ADSs to estimate failure rates and potential safety risks. Similarly, publicly available reports from ADS companies such as Waymo [ 20 ] have provided valuable insights into the circumstances that necessitate human intervention in autonomous driving. Disengagements during remote driving, in which a SD overwrites the inputs of the RD, have a comparable significance. For remote driving purposes