Over 60,000 km in a year: remotely collecting large-volume high-quality data from a logistics truck

Berger, Christian; Karsolia, Arpit; Giaimo, Federico; Benderius, Ola · 2022 · Crossref

DOI: 10.1007/s42452-022-05159-w

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This paper addresses the challenge of collecting large-volume, high-quality data from commercial freight trucks to support the development of SAE Level 4 autonomous driving systems. While previous autonomous driving research relied on specialized test vehicles or low-volume data loggers, there is a critical need for data reflecting real-world logistics operations. The authors present an end-to-end data logging solution installed in a truck operated daily by a logistics company, enabling remote, over-the-air data collection without physical access to the vehicle. The primary research questions focus on designing a reliable system architecture for cellular-only data exchange and reporting lessons learned from one year of operation. The methodology employs design science to create a fail-safe system architecture that avoids single points of failure. The hardware setup includes two independent computing nodes ("Eagle" and "Apollo") connected to five cameras, two GNSS-IMU sensors, and six vehicle CAN buses. To ensure reliability, the system uses redundant 3G/4G modems, separate power supplies, and an auxiliary lithium-ion battery that allows data upload after the engine is shut off. The software stack, running on Arch Linux with Docker microservices, handles lossless video compression using GPU acceleration and continuously uploads data to a cloud storage system. The system was deployed in October 2019 and monitored remotely via encrypted channels. During the first year of operation, the truck traveled over 62,800 km, with data logged on 193 days, resulting in more than 4.5 TB of uploaded data. The system successfully captured over 106 million video frames and maintained high system uptime. The authors validated the data quality by analyzing longitudinal acceleration to identify harsh braking events (defined as ≤ −0.5 G). This analysis identified 23 potential events, of which five were confirmed as genuine harsh braking maneuvers caused by traffic interactions, such as lane merging, while eight were false positives. The system demonstrated stable performance, with median CPU loads well within operational limits and consistent GNSS satellite visibility. The study concludes that remote, large-volume data logging from operational commercial vehicles is feasible and valuable for training AI/ML algorithms for autonomous driving. The proposed architecture successfully met functional requirements for secure, wireless access and non-functional requirements for redundancy and lossless data handling. The authors highlight that this approach provides unique insights into real-world driving behaviors and critical events that are difficult to replicate in controlled test environments. The paper also documents hardware issues, such as power supply failures, and software fixes, offering practical lessons for future deployments. This work pioneers the use of daily-operated trucks for data collection, bridging the gap between experimental prototypes and commercially viable autonomous freight solutions.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• driverless ads

Information type

What kind of knowledge this paper contributes, grouped by family — independent of topic (what it is about) and method (how it was studied).

• Methodological Resource: dataset resource, tool software