Pith. sign in

REVIEW 3 major objections 5 minor 1 cited by

Teleoperating Autonomous Vehicles over Commercial 5G Networks: Are We There Yet?

T0 review · 3 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read Today's commercial 5G networks, as deployed, cannot carry the full sensor uplink that teleoperated driving needs.

desk verdict Solid field benchmark for AV teleoperation over commercial 5G, but the uplink latency claims rest on a USB-tethered replay proxy that the paper never validates. read the letter →

arxiv 2507.20438 v1 pith:EKDWTI25 submitted 2025-07-27 cs.NI cs.OH

classification cs.NIcs.OH
keywords 5GteleoperateddrivingautonomousvehiclesuplinklatencyPHYlayerhandoverLiDARstreamingQoEmeasurement
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper asks whether today's commercial 5G networks can support teleoperated driving, and answers that they cannot yet carry the full sensor picture. It streams real camera and LiDAR data over a standalone 5G carrier in urban driving loops, splitting latency into per-frame network delay versus queueing and processing delay. The central numbers: a single raw front camera has a median per-frame network delay of 73.5 ms against a 45 ms network-level target, and 29.2% of frames miss the 100 ms end-to-end deadline. Streaming multiple cameras plus high-resolution LiDAR is nearly impossible without aggressive compression, which in turn degrades video quality and downstream object detection. The paper's contribution is a cross-layer diagnosis locating the bottleneck in 5G uplink asymmetry, retransmissions, and handover behavior rather than in the streaming application alone.

What carries the argument

The machinery is a per-frame QoE decomposition of uplink sensor streaming, splitting end-to-end delay into per-frame network delay (first packet sent to last packet received) versus queueing and encoding/decoding delay, and then cross-correlating per-frame network delay time series with PHY-layer traces—CQI, MCS, BLER, resource-block allocation, and handover events. This decomposition is what lets the paper assign deadline violations to specific radio behaviors rather than to the application, and it drives the quantitative attributions (CQI, BLER, handover impacts) that form the feasibility verdict.

What would settle it

Replay the same driving loops while streaming live sensor data from the AV's onboard computer over an integrated 5G modem, not through a USB-tethered phone, and compare the per-frame network delay CDF. If the median drops below 45 ms and fewer than about 5% of frames miss the 100 ms end-to-end deadline, the paper's infeasibility verdict for today's commercial 5G would be overturned by its own measurement standard.

Watch

Extended reading notes

Core claim

The paper's central claim is a feasibility verdict: commercial 5G networks as they operate today cannot support teleoperated driving that requires full sensor uploads. Even the easiest case—a single raw front-camera feed over a standalone 5G carrier—shows a median per-frame network delay of 73.5 ms against a 45 ms network-level target, and 29.2% of frames miss the 100 ms end-to-end deadline. Compressing the video (H.264/H.265 or VP8/VP9) brings most frames under the deadlines, but the tail remains dangerous, and merged multi-camera streams plus a 64-beam LiDAR stream are nearly impossible without aggressive downsampling. The paper also attributes the failures to specific 5G radio behaviors: poor channel conditions raise average per-frame delay by up to 92.5%, retransmissions by about 55.7%, and ping-pong handovers during turns by 56–85%. It concludes that application-layer adaptation such as WebRTC's reacts too slowly to 5G PHY dynamics, so the fix must come from co-design of the radio, edge cloud, and streaming application.

Load-bearing premise

The whole feasibility verdict rests on treating replay of pre-recorded sensor data through USB-tethered phones as equivalent to the AV's real integrated radio path, since the paper never validates that proxy against live on-vehicle streaming.

Editorial extensions

If this is right

  • A teleoperation service built on a single compressed camera feed can work much of the time, but its worst moments—clustered tail-latency events during handovers and poor radio conditions—are exactly when a safety-critical intervention may be needed.
  • Full situational awareness (multiple cameras plus 64- or 128-beam LiDAR) is beyond today's commercial 5G uplink capacity, so practical teleoperation designs must either aggressively reduce sensor data or restrict the operational domain.
  • Compression lowers delay but degrades perceptual quality and downstream object detection nonlinearly, so latency and perception quality cannot be treated as independent knobs.
  • WebRTC-style congestion control responds seconds after the 5G PHY layer has already degraded, so application-layer-only adaptation cannot prevent the queuing spikes; 5G-aware or cross-layer feedback would be needed.
  • Operator-level switching (using one carrier at a time) is a more promising near-term mitigation than packet-level splitting across carriers, which allows a single bad channel to delay an entire frame.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • Beyond the paper: the verdict applies to commercial 5G configured for best-effort mobile internet; a network slice with dedicated uplink resources or a URLLC profile could pass the same per-frame deadline test even though today's default network does not.
  • Beyond the paper: the handover results suggest a trajectory-aware handover trigger—suppressing ping-pong handovers when the vehicle is turning—would likely reduce tail-latency violations more than adding bandwidth, since the paper shows handover-induced delay increases of 56–85%.
  • Beyond the paper: the per-frame QoE metrics could be reused as a continuous safety monitor in a production teleoperation system, flagging moments when the frame delay distribution shifts into the tail rather than relying on average latency.
  • Beyond the paper: because the sensor data was replayed rather than streamed live from the vehicle's integrated radio path, a natural next experiment is a head-to-head comparison of tethered-phone versus onboard-modem uplink latency on the same loops; a difference of even a few tens of milliseconds would shift the single-camera feasibility boundary.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper reports a six-month field measurement campaign in Minneapolis that evaluates whether commercial 5G networks can support teleoperated driving. The testbed streams camera, LiDAR, and command-and-control data from a research AV to an AWS edge server over commercial 5G, with emphasis on T-Mobile's 5G-SA network, while collecting PHY-layer RAN metrics using the XCAL tool. The authors define per-frame latency and quality metrics, compare against 5GAA latency thresholds (100 ms application-level UL, 45 ms network-level UL), and analyze how CQI, BLER, handovers, and resource-block allocation affect per-frame delay. They also study WebRTC and RTSP adaptation behavior, multi-AV resource contention, and multi-operator switching. The central conclusion is that single-camera streaming is feasible in most scenarios but has unsafe tail latency, whereas streaming multiple cameras plus LiDAR is effectively infeasible on today's commercial networks.

Significance. If the results hold, this is a valuable and unusually comprehensive real-world data point: it combines application-level sensor streaming with synchronized PHY-layer logs, uses externally defined 5GAA thresholds rather than fitted models, and evaluates the downstream AI-task impact of compression. The campaign scale (approximately 70 loops, 100s of GB, 6 months) and the per-frame QoE metrics are genuine strengths. However, the central latency claims currently rest on an unvalidated radio-path proxy, and at least one headline quantitative comparison is internally inconsistent. These issues must be resolved before the feasibility verdict can be taken as established.

major comments (3)
  1. [Section 4.1 (Data Collection Approach); Figs. 6, 9, 12-14, 17] The uplink measurements are made over a proxy path: the sensor data were recorded during a 1748-km drive and later replayed from the on-board computer through USB-tethered Samsung Galaxy S21 Ultra smartphones, because the XCAL logging tool only supports Samsung phones. The manuscript never validates that this USB-tethered path reproduces the per-frame latency behavior of the AV's own integrated radio path, and it never quantifies the additional USB serialization/queueing hop. This is load-bearing for the central feasibility verdict, since the headline numbers -- the single-camera median per-frame network delay of 73.5 ms, the 29.2% of frames exceeding 100 ms E2E, and the LiDAR median delays of 2-6 s -- are all measured over this proxy. The authors should either run a validation experiment comparing the tethered-phone path against the AV's native radio path, or explicitly reposition the paper's latency claims as applying to a USB-tethered smartphone-based radio path and analyze how the unmodeled hop could shift the conclusions.
  2. [Section 1 vs. Section 6.2 (CQI impact)] The key-findings bullet in Section 1 states that per-frame delay increases by 'about 48%' when CQI goes from good to poor, while Section 6.2 reports a '92.5% increase' (770 ms vs 400 ms) for the same qualitative comparison. These two statements contradict each other. The authors should align the summary with the body, or, if the two numbers describe different experiments, identify the experiment and conditions for each.
  3. [Section 6.2 (Handover analysis) and Fig. 14] The 86.04% during-handover delay increase and 7.83% post-handover improvement are reported as single aggregate numbers, but the section does not state how many handover events underlie them or provide confidence intervals. Given that the handover impact is one of the paper's main safety-relevant conclusions, the quantitative claim needs more statistical detail before it can be assessed.
minor comments (5)
  1. [Section 5.1 (Per-Frame Network Delay)] The phrase '28 ms higher than the maximum 5G network delay threshold' is awkward because Table 1 gives a range (40-45 ms); the authors should state the specific target value used and justify why the per-frame network delay should be compared directly to the 5G network-level latency target.
  2. [Section 6.2 (Ping-pong HOs)] There is a typo: 'pong-pong HOs' should be 'ping-pong HOs'.
  3. [Table 4] The left camera's raw data rate is shown as '37.749 * 30' with an unexplained asterisk; either remove the asterisk or explain what it denotes.
  4. [Appendix 10.3 (Command & Control)] The C&C experiments replay pre-recorded Logitech simulator commands over gRPC rather than performing live teleoperation; this should be disclosed in the main text where Section 5.4 reports C&C delay results.
  5. [Section 1 / Abstract] The data-collection path through USB-tethered smartphones is not disclosed in the abstract or introduction; given its importance for interpreting all UL latency results, it should be mentioned prominently.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the feasibility verdict is a measurement study compared against external 5GAA thresholds, and the self-citations are contextual rather than load-bearing.

full rationale

This paper is a field measurement study, not a derivation from fitted parameters, so the circularity burden is naturally low. The central feasibility claims are comparisons of measured per-frame UL delays and throughputs against externally specified 5GAA requirements (Tables 1 and 2): for example, the median per-frame network delay of 73.5 ms is compared with the 40-45 ms network-level target, and LiDAR throughput requirements (277-307 Mbps) are compared with measured UL PHY throughput (77.7 Mbps for T-Mobile). These thresholds come from 5GAA documents and sensor specifications, not from the paper's own conclusions, so the comparison is not self-referential. The new per-frame QoE metrics in Section 4.2 are operational definitions of quantities that are then measured; they are not defined in terms of the feasibility outcome. The CQI/BLER/handover analyses in Section 6 are empirical correlations between independent RAN metrics and measured delays, not fitted predictors disguised as findings. The paper does cite the authors' own prior work, e.g., [21] and [72] for the claim that T-Mobile is the only carrier with a primary 5G-SA deployment, and [29, 43] for latency measurement methodology; however, these citations support background or methodological context, are externally checkable, and are not the load-bearing step that produces the feasibility verdict. The replay and USB-tethered methodology described in Section 4.1 is a legitimate validity limitation, because the sensor data were recorded over 1748 km and replayed through Samsung Galaxy S21 Ultra smartphones rather than streamed live from the AV's integrated radio path; this affects whether the measured latencies generalize to a real AV, but it is not circularity, since the measurements are not constructed from or equivalent to the paper's conclusions. Similarly, the inconsistency between the abstract's '48%' CQI delay increase and Section 6.2's '92.5%' is a reporting/reproducibility problem, not a circular derivation. No step in the paper reduces, by construction or by self-citation chain, to its own inputs.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The central claim rests on external 5GAA requirements, on the validity of a replay-and-tether testbed, and on the fidelity of RAN logging; no free parameters are fitted and no new entities are postulated.

assumptions (5)
  • domain assumption 5GAA teleoperation latency thresholds (100 ms UL / 20 ms DL application level; 45 ms / 15 ms network level) define feasibility.
    The paper uses these thresholds in Tables 1-2 and throughout Sections 5-8 to judge whether per-frame delays are acceptable; they come from 5GAA references [10,11,12].
  • domain assumption Replaying pre-recorded sensor data from a laptop via USB-tethered 5G smartphones is a valid proxy for live AV sensor streaming over 5G.
    Section 4.1 Data Collection Approach; every UL latency measurement in Sections 5-7 inherits this assumption.
  • domain assumption RAN parameters captured by XCAL on the tethered smartphones are representative of the radio conditions experienced by the AV data traffic.
    Section 6 cross-correlation analysis assumes logged CQI/BLER/HO/RB events correspond to the streaming path.
  • domain assumption T-Mobile 5G-SA results generalize to other commercial 5G networks.
    The in-depth analysis is limited to T-Mobile, selected as the only standalone 5G carrier available; AT&T and Verizon appear only in throughput baselines.
  • domain assumption Standard network measurement tools (iPerf3, Wireshark, Accuver XCAL, NTP sync) provide unbiased timing and RAN information.
    All delay and PHY-layer measurements rely on the accuracy of these tools, described in Sections 4.1 and 4.2.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Teleoperating Autonomous Vehicles over Commercial 5G Networks: Are We There Yet?." pith.science (2026). https://pith.science/paper/EKDWTI25

@misc{pith2026250720438,
  author       = {Pith},
  title        = {Pith review of: Teleoperating Autonomous Vehicles over Commercial 5G Networks: Are We There Yet?},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/EKDWTI25}},
  note         = {Machine review of arXiv:2507.20438}
}
read the original abstract

Remote driving, or teleoperating Autonomous Vehicles (AVs), is a key application that emerging 5G networks aim to support. In this paper, we conduct a systematic feasibility study of AV teleoperation over commercial 5G networks from both cross-layer and end-to-end (E2E) perspectives. Given the critical importance of timely delivery of sensor data, such as camera and LiDAR data, for AV teleoperation, we focus in particular on the performance of uplink sensor data delivery. We analyze the impacts of Physical Layer (PHY layer) 5G radio network factors, including channel conditions, radio resource allocation, and Handovers (HOs), on E2E latency performance. We also examine the impacts of 5G networks on the performance of upper-layer protocols and E2E application Quality-of-Experience (QoE) adaptation mechanisms used for real-time sensor data delivery, such as Real-Time Streaming Protocol (RTSP) and Web Real Time Communication (WebRTC). Our study reveals the challenges posed by today's 5G networks and the limitations of existing sensor data streaming mechanisms. The insights gained will help inform the co-design of future-generation wireless networks, edge cloud systems, and applications to overcome the low-latency barriers in AV teleoperation.

Figures

Figures reproduced from arXiv: 2507.20438 by the authors.

Figure 1
Figure 1. Experimental Setup, Tools, & Streaming System. [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 4
Figure 4. PHY DL & UL Latency [PITH_FULL_IMAGE:figures/full_fig_p004_4.png] view at source ↗
Figure 5
Figure 5. Illustration of the Delay QoE Metrics in Cam￾era/LiDAR Streaming [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figures from the paper (14 more)
Figure 6
Figure 6. Figure 6: QoE Metrics When Streaming the Raw Front-Central Camera – the Most Crucial Data Feed Over TM 5G 0 50 100 150 200 Per-frame Total Delay (ms) 0.0 0.2 0.4 0.6 0.8 1.0 CDF 0.08 0.75 0.24 0.97 h264 h265 (a) [UL] Per-Frame Total Delay for I Frames (b) [UL] Per-Frame Net￾work…
Figure 8
Figure 8. Figure 8: Front-Camera Streaming VP8 and VP9 Performance with WebRTC delay effect (see Fig. 6b top figure). This is because the AV’s camera data begins to queue in the UL buffer. (3) The queuing is triggered by a domino effect due to the 5G network’s inabil￾ity to transmit data …
Figure 10
Figure 10. Figure 10: [DL] C&C vs. [UL] Video Frame Delay. infeasible to stream high-resolution LiDAR data in real-time over 5G networks. We omit the results for streaming 128 beams and multiple LiDAR streams, as they will obviously perform worse. With Draco, although we can significantly …
Figure 11
Figure 11. Figure 11: 5G Impact on Per-Frame (UL) Network Delay for Teleoperation. Using RTSP to Stream: (a) Single (front) Camera, and (b) Merged (front, left, & right) Cameras find that 64.29% and 37.20% of the C&C messages were de￾livered within the application and network requirements,…
Figure 14
Figure 14. Figure 14: HOs Impact on Per-frame Net￾work Delay several HOs results in much more severe delays than the effect of a single HO during the same period – this is a key point that we explore further in §6.2. (3) BLERs also increase Per-frame network delay, although the effects are…
Figure 15
Figure 15. Figure 15: Full Drive Loop, showing the density of PCIs. [PITH_FULL_IMAGE:figures/full_fig_p011_15.png]
Figure 16
Figure 16. Figure 16: Ping-Pong HOs occurring while driving in a [PITH_FULL_IMAGE:figures/full_fig_p011_16.png]
Figure 18
Figure 18. Figure 18: WebRTC’s Bitrate Choices vs Avail￾able Throughput. 25 50 Single-user 25 50 PHY-layer resource blocks Multi-user(UE1) 0 10 20 30 time (minutes) 25 50 Multi-user(UE2) (a) RBs Distribution 200 400 600 1000 1600 Per-frame total delay(s) 0.0 0.2 0.4 0.6 0.8 1.0 CDF 269 661…
Figure 17
Figure 17. Figure 17: Cross-Layer Analysis of WebRTC Single Camera [PITH_FULL_IMAGE:figures/full_fig_p012_17.png]
Figure 20
Figure 20. Figure 20: Multi-Path Streaming for Single Video In [PITH_FULL_IMAGE:figures/full_fig_p013_20.png]
Figure 22
Figure 22. Figure 22: WebRTC QoE for merged-video streaming over a 5G network to the remote vehicle control station, enabling comprehensive situational awareness for teleopera￾tion. The results from our experiments with merged-video streaming using VP8 codec are illustrated in [PITH_FULL_…
Figure 21
Figure 21. Figure 21: Merged Frames from Front Left, Front, and [PITH_FULL_IMAGE:figures/full_fig_p016_21.png]
Figure 23
Figure 23. Figure 23: Object Detection with Different Data Modalities and Compression Qualities. [PITH_FULL_IMAGE:figures/full_fig_p017_23.png]
Figure 25
Figure 25. Figure 25: Video Compression Codec Effect on ML De [PITH_FULL_IMAGE:figures/full_fig_p017_25.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Bilateral Trade Under Heavy-Tailed Valuations: Minimax Regret with Infinite Variance

    stat.ML 2026-03 unverdicted novelty 7.0 of 10

    Under bounded density and finite p-th moments with p in (1,2), contextual bilateral trade has exact minimax regret rate T to the power 1 minus 2β(p-1)/(βp + d(p-1)).

Reference graph

Works this paper leans on

81 extracted references · 69 canonical work pages · cited by 1 Pith paper

  1. [1]

    [n. d.]. 5G network slicing. https://en.wikipedia.org/wiki/5G_ network_slicing

  2. [2]

    [n. d.]. Amazon Web Services (AWS). https://aws.amazon.com/

  3. [3]

    [n. d.]. Sizing the Solution. https://infohub.delltechnologies.com/en- us/l/computer-vision-3d-flow-and-function-ai-with-lidar/sizing- the-solution-53/6/#:~:text=Network%20sizing%20is%20extremely% 20important,s%20and%2050%20Mb%2Fs

  4. [4]

    Taxonomy and Definitions for Terms Related to Driving Au- tomation Systems for On-Road Motor Vehicles, SAE International Recommended Practice Standard J3016-2018

    2018. Taxonomy and Definitions for Terms Related to Driving Au- tomation Systems for On-Road Motor Vehicles, SAE International Recommended Practice Standard J3016-2018. https://www.sae.org/ standards/content/j3016_201806/

  5. [5]

    World’s First Remotely-Controlled 5G Car To Make History At Goodwood Festival of Speed

    2019. World’s First Remotely-Controlled 5G Car To Make History At Goodwood Festival of Speed. https://news.samsung.com/uk/worlds- first-remotely-controlled-5g-car-to-make-history-at-goodwood- festival-of-speed

  6. [6]

    Startup vay’s Autonomy Workaround: Teledrivers to Operate Cars from Remote Location

    2021. Startup vay’s Autonomy Workaround: Teledrivers to Operate Cars from Remote Location. https://www.caranddriver.com/news/ a37648114/vay-autonomous-teledriver-startup/

  7. [7]

    This Driverless Car-Sharing Service uses Remote Human “Pilots”, not AI

    2021. This Driverless Car-Sharing Service uses Remote Human “Pilots”, not AI. https://www.fastcompany.com/90653650/halo-driverless-car- sharing-service

  8. [8]

    Accuver XCAL

    2022. Accuver XCAL. https://www.accuver.com/sub/products/view. php?idx=6&ckattempt=2

Show all 81 references
  1. [9]

    3GPP. 2020. 5G; NR; Physical layer procedures for data (3GPP TS 38.214 version 16.2.0 Release 16). https://www.etsi.org/deliver/etsi_ts/ 138200_138299/138214/16.02.00_60/ts_138214v160200p.pdf

  2. [10]

    5GAA. 2021. C-V2X Use Cases and Service Level Requirements Volume II. https://5gaa.org/c-v2x-use-cases-and-service-level-requirements- volume-ii/, Last accessed: Sept 20, 2024

  3. [11]

    5GAA. 2021. Tele-Operated Driving (ToD): System Requirements Analysis and Architecture. https://5gaa.org/tele-operated-driving- tod-system-requirements-analysis-and-architecture/, Last accessed: Sept 20, 2024

  4. [12]

    5GAA. 2024. 5G Automotive Association. https://5gaa.org/, Last accessed: Sept 20, 2024

  5. [13]

    5GCroCo. 2024. 5GCroCo: 5G for Cooperative, Connected and Auto- mated Mobility. https://5gcroco.eu/, Last accessed: Sept 20, 2024

  6. [14]

    Manzoor Ahmed, Salman Raza, Muhammad Ayzed Mirza, Abdul Aziz, Manzoor Ahmed Khan, Wali Ullah Khan, Jianbo Li, and Zhu Han

  7. [15]

    aler9 and github contributors. [n. d.]. rtsp simple server. https: //github.com/aler9/rtsp-simple-server Accessed February 2023

  8. [16]

    Rory Bennett, Reyn Kapp, Theunis R Botha, and Schalk Els. 2020. Influence of wireless communication transport latencies and dropped packages on vehicle stability with an offsite steering controller. IET Intelligent Transport Systems 14, 7 (2020), 783–791

  9. [17]

    Jon Brodkin. 2023. After robotaxi dragged pedestrian 20 feet, Cruise founder and CEO resigns. https://arstechnica.com/tech- policy/2023/11/after-robotaxi-dragged-pedestrian-20-feet-cruise- 13 Rostand A. K. Fezeu, Jason Carpenter et al. founder-and-ceo-resigns/

  10. [18]

    Martin Buehler, Karl Iagnemma, and Sanjiv Singh. 2007. The 2005 DARPA Grand Challenge: The Great Robot Race (1st ed.). Springer Publishing Company, Incorporated

  11. [19]

    Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom

    Holger Caesar, Varun Bankiti, Alex H. Lang, Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, and Oscar Beijbom. 2019. nuScenes: A multimodal dataset for autonomous driving. arXiv preprint arXiv:1903.11027 (2019)

  12. [20]

    Ricardo Cano. 2024. One crash set off a new era for self-driving cars in S.F. Here’s a complete look at what happened. https://www.sfchronicle. com/projects/2024/cruise-sf-collision-timeline/

  13. [21]

    Jason Carpenter, Wei Ye, Feng Qian, and Zhi-Li Zhang. 2023. Multi- Modal Vehicle Data Delivery via Commercial 5G Mobile Networks: An Initial Study. In 2023 IEEE 43rd International Conference on Distributed Computing Systems Workshops (ICDCSW). IEEE, 157–162

  14. [22]

    Yilun Chen, Zhiding Yu, Yukang Chen, Shiyi Lan, Animashree Anand- kumar, Jiaya Jia, and Jose Alvarez. 2023. FocalFormer3D : Focusing on Hard Instance for 3D Object Detection. arXiv:2308.04556 [cs.CV]

  15. [23]

    A. Davies. 2018. Self-driving cars have a secret weapon : Remote control. https://www.wired.com/story/phantom-teleops/

  16. [24]

    A. Davies. 2019. The war to remotely control self-driving cars heats up. https://www.wired.com/story/designated-driver-teleoperations- self-driving-cars/

  17. [25]

    Jos den Ouden, Victor Ho, Tijs van der Smagt, Geerd Kakes, Simon Rommel, Igor Passchier, Jakub Juza, and Idelfonso Tafur Monroy. 2022. Design and Evaluation of Remote Driving Architecture on 4G and 5G Mobile Networks. Frontiers in Future Transportation 2 (2022). https: //doi.o...

  18. [26]

    Google Developers. 2024. https://webrtc.org/ accessed Nov 2024

  19. [27]

    Mohyeldin Eiman. 2020. Minimum Technical Perfor- mance Requirements for IMT-2020 radio interface(s). https://www.itu.int/en/ITU-R/study-groups/rsg5/rwp5d/imt- 2020/Documents/S01-1_Requirements%20for%20IMT-2020_Rev.pdf

  20. [28]

    Rostand A. K. Fezeu, Jason Carpenter, Claudio Fiandrino, Eman Ramadan, Wei Ye, Joerg Widmer, Feng Qian, and Zhi-Li Zhang

  21. [29]

    Rostand A. K. Fezeu, Eman Ramadan, Wei Ye, Benjamin Minneci, Jack Xie, Arvind Narayanan, Ahmad Hassan, Feng Qian, Zhi-Li Zhang, Jaideep Chandrashekar, and Myungjin Lee. 2023. An In-Depth Mea- surement Analysis of 5G mmWave PHY Latency and Its Impact on End-to-End Delay. In Pas...

  22. [30]

    Claudio Fiandrino and et al. 2022. Uncovering 5G performance on public transit systems with an app-based measurement study. In Pro- ceedings of the 25th International ACM Conference on Modeling Analysis and Simulation of Wireless and Mobile Systems . 65–73

  23. [31]

    Jonny Kong, Phuc Dinh, Jiayi Meng, Y

    Moinak Ghoshal, Imran Khan, Z. Jonny Kong, Phuc Dinh, Jiayi Meng, Y. Charlie Hu, and Dimitrios Koutsonikolas. 2023. Performance of Cellular Networks on the Wheels. In Proceedings of the 2023 ACM on Internet Measurement Conference (IMC ’23). Association for Computing Machinery,...

  24. [32]

    Jonny Kong, Qiang Xu, Zixiao Lu, Shivang Aggar- wal, Imran Khan, Yuanjie Li, Y

    Moinak Ghoshal, Z. Jonny Kong, Qiang Xu, Zixiao Lu, Shivang Aggar- wal, Imran Khan, Yuanjie Li, Y. Charlie Hu, and Dimitrios Koutsoniko- las. 2022. An In-Depth Study of Uplink Performance of 5G MmWave Networks. In Proceedings of the ACM SIGCOMM Workshop on 5G and Beyond Networ...

  25. [33]

    Google. [n. d.]. GitHub - google/draco: Draco library. https://github. com/google/draco. [Accessed 13-06-2024]

  26. [34]

    Google. n.d.. gRPC: A high performance, open-source universal RPC framework. https://grpc.io/. https://grpc.io

  27. [35]

    M. Harris. 2018. CES 2018: Phantom Auto Demonstrates First Remote- Controlled Car on Public Roads. https://spectrum.ieee.org/ces-2018- phantom-auto-demonstrates-first-remotecontrolled-car-on-public- roads. IEEE Spectrum, Piscataway, NJ, USA

  28. [36]

    Morley Mao, Feng Qian, and Zhi-Li Zhang

    Ahmad Hassan, Arvind Narayanan, Anlan Zhang, Wei Ye, Ruiyang Zhu, Shuowei Jin, Jason Carpenter, Z. Morley Mao, Feng Qian, and Zhi-Li Zhang. 2022. Vivisecting Mobility Management in 5G Cellular Networks. In Proc. of ACM SIGCOMM . 86–100. https://doi.org/10. 1145/3544216.3544217

  29. [37]

    ANDREW J. HAWKINS. 2022. Cruise’s driverless robotaxis are accept- ing passengers in Phoenix and Austin. https://www.theverge.com/ 2022/12/20/23518833/cruise-driverless-taxi-austin-phoenix-waitlist

  30. [38]

    ANDREW J. HAWKINS. 2022. Waymo’s driverless ve- hicles are picking up passengers in downtown Phoenix. https://www.theverge.com/2022/8/29/23323593/waymo-driverless- vehicles-passengers-downtown-phoenix

  31. [39]

    Kuhn, Goran Petrovic, and Eckehard Steinbach

    Markus Hofbauer, Christopher B. Kuhn, Goran Petrovic, and Eckehard Steinbach. 2020. TELECARLA: An Open Source Extension of the CARLA Simulator for Teleoperated Driving Research Using Off-the- Shelf Components. In 31st IEEE Intelligent Vehicles Symposium 2020 (IV). IEEE, Las Ve...

  32. [40]

    Lila Huang, Shenlong Wang, Kelvin Wong, Jerry Liu, and Raquel Urtasun. 2020. Octsqueeze: Octree-structured entropy model for lidar compression. In Proceedings of the IEEE/CVF conference on computer vision and pattern recognition . 1313–1323

  33. [41]

    David Ingram. 2023. Two companies race to deploy rob- otaxis in San Francisco. The city wants them to hit the brakes. https://www.nbcnews.com/tech/tech-news/san-francisco- looks-hit-brakes-self-driving-cars-rcna66204

  34. [42]

    iperf3 community. 2023. iPerf3. https://iperf.fr/iperf-download.php

  35. [43]

    Rostand A. K. Fezeu and et al. 2024. Unveiling the 5G Mid-Band Land- scape: From Network Deployment to Performance and Application QoE. In Proceedings of the ACM SIGCOMM 2024 Conference

  36. [44]

    Rostand A. K. Fezeu, Claudio Fiandrino, Eman Ramadan, Jason Car- penter, Daqing Chen, Yiling Tan, Feng Qian, Joerg Widmer, and Zhi-Li Zhang. 2024. Roaming across the European Union in the 5G Era: Performance, Challenges, and Opportunities. In Proc. of IEEE INFO- COM. Available...

  37. [45]

    Riley Kaminer. 2024. With new office, Guident zooms into the future of autonomous driving. https://refreshmiami.com/news/with-new- office-guident-zooms-into-the-future-of-autonomous-driving/

  38. [46]

    Philip Koopman and Michael Wagner. 2016. Challenges in Au- tonomous Vehicle Testing and Validation. SAE International Journal of Transportation Safety 4, 1 (April 2016), 15–24. https://doi.org/10. 4271/2016-01-0128

  39. [47]

    Liu and C

    Y. Liu and C. Peng. 2023. A Close Look at 5G in the Wild: Unrealized Potentials and Implications. In Proc. of IEEE INFOCOM . 1–10

  40. [48]

    Cade Metz, Jason Henry, Ben Laffin Laffin, Rebecca Lieber- man, and Yiwen Lu. 2024. How Self-Driving Cars Get Help From Humans Hundreds of Miles Away. New York Times (2024). https://www.nytimes.com/interactive/2024/09/03/technology/ zoox-self-driving-cars-remote-control.html

  41. [49]

    Arvind Narayanan, Eman Ramadan, Jason Carpenter, Qingxu Liu, Yu Liu, Feng Qian, and Zhi-Li Zhang. 2020. A First Look at Commercial 5G Performance on Smartphones. In Proc. of The Web Conference . 894–905. 14 Teleoperating A Vs Over Commercial 5G Networks: Are We There Yet?

  42. [50]

    Arvind Narayanan, Eman Ramadan, Rishabh Mehta, Xinyue Hu, Qingxu Liu, Rostand AK Fezeu, Udhaya Kumar Dayalan, Saurabh Verma, Peiqi Ji, Tao Li, Feng Qian, and Zhi-Li Zhang. 2020. Lumos5G: Mapping and predicting commercial mmWave 5G throughput. InProc. of the ACM Internet Measur...

  43. [51]

    Arvind Narayanan, Muhammad Iqbal Rochman, Ahmad Hassan, Bariq S Firmansyah, Vanlin Sathya, Monisha Ghosh, Feng Qian, and Zhi-Li Zhang. 2022. A comparative measurement study of commercial 5G mmwave deployments. In IEEE INFOCOM 2022. IEEE, 800–809

  44. [52]

    Arvind Narayanan, Xumiao Zhang, Ruiyang Zhu, Ahmad Hassan, Shuowei Jin, Xiao Zhu, Xiaoxuan Zhang, Denis Rybkin, Zhengx- uan Yang, Zhuoqing Morley Mao, Feng Qian, and Zhi-Li Zhang

  45. [53]

    Yunzhe Ni, Zhilong Zheng, Xianshang Lin, Fengyu Gao, Xuan Zeng, Yirui Liu, Tao Xu, Hua Wang, Zhidong Zhang, Senlang Du, et al

  46. [54]

    Associated Press. [n. d.]. Driverless taxis are coming to the streets of San Francisco. NPR Technology, https://www.npr.org/2022/06/03/ 1102922330/driverless-self-driving-taxis-san-francisco-gm-cruise, June 3, 2022. Last accessed: June 8, 2022

  47. [55]

    Sreenan, and Jason J

    Darijo Raca, Dylan Leahy, Cormac J. Sreenan, and Jason J. Quinlan

  48. [56]

    Eman Ramadan, Arvind Narayanan, Udhaya Kumar Dayalan, Rostand A. K. Fezeu, Feng Qian, and Zhi-Li Zhang. 2021. Case for 5G-Aware Video Streaming Applications. In Proc. of the 5G-MeMU . 27–34

  49. [57]

    In Proceedings of the ACM SIGCOMM 2023 Conference

    CellFusion: Multipath Vehicle-to-Cloud Video Streaming with Network Coding in the Wild. In Proceedings of the ACM SIGCOMM 2023 Conference. 668–683

  50. [58]

    RoboAuto. 2024. RoboAuto. https://roboauto.tech/, Last accessed: Sept 20, 2024

  51. [59]

    Ibrahim, and William Payne

    Muhammad Iqbal Rochman, Vanlin Sathya, Norlen Nunez, Damian Fernandez, Monisha Ghosh, Ahmed S. Ibrahim, and William Payne

  52. [60]

    Muhammad Iqbal Rochman, Wei Ye, Zhi-Li Zhang, and Monisha Ghosh. 2024. A Comprehensive Real-World Evaluation of 5G Im- provements over 4G in Low-and Mid-Bands. IEEE DySPAN’24 (2024)

  53. [61]

    EMMA ROTH. 2023. San Francisco wants to slow robotaxi rollout over blocked traffic and false 911 calls. https://www.theverge.com/ 2023/1/29/23576422/san-francisco-cruise-waymo-robotaxi-rollout

  54. [62]

    Joseph Redmon. [n. d.]. YOLO: Real-Time Object Detection — pjred- die.com. https://pjreddie.com/darknet/yolo/. [Accessed 14-06-2024]

  55. [63]

    SAE International. 2018. Surface vehicle. SAE International

  56. [64]

    Alcaraz-Calero, and Jose Garcia- Rodriguez

    Javier Saez-Perez, Qi Wang, Jose M. Alcaraz-Calero, and Jose Garcia- Rodriguez. 2023. Design, Implementation, and Empirical Validation of a Framework for Remote Car Driving Using a Commercial Mobile Network. Sensors 23, 3 (2023). https://doi.org/10.3390/s23031671

  57. [65]

    A Comparison Study of Cellular Deployments in Chicago and Miami Using Apps on Smartphones. InProc. of ACM WiNTECH. 61–68. https://doi.org/10.1145/3477086.3480843

  58. [66]

    Gaurav Sharma and Rajesh Rajamani. 2024. Teleoperation Enhancement for Autonomous Vehicles Using Estimation Based Predictive Display. https://drive.google.com/file/d/ 1HBbEbzKttGW4TQPHQFWeg39CNALC_351/view?usp=drive_link, Last accessed: Sept 20, 2024

  59. [67]

    Wireshark Team. [n. d.]. Wireshark. https://www.wireshark.org/

  60. [68]

    SAE International. 2014. Automated Driving: Levels of Driving Au- tomation are Defined in New SAE International Standard J3016. SAE International Troy, MI

  61. [69]

    Justin Uberti, Stefan Holmer, Magnus Flodman, Danny Hong, and Jonathan Lennox. 2021. RTP Payload Format for VP9 Video . Internet- Draft draft-ietf-payload-vp9-16. Internet Engineering Task Force. https://datatracker.ietf.org/doc/draft-ietf-payload-vp9/16/ Work in Progress

  62. [70]

    Paul Wilkins, Yaowu Xu, Lou Quillio, James Bankoski, Janne Salonen, and John Koleszar. 2011. VP8 Data Format and Decoding Guide. RFC

  63. [71]

    Andreas Schimpe, Johannes Feiler, Simon Hoffmann, Domagoj Ma- jstorović, and Frank Diermeyer. 2022. Open Source Software for Teleoperated Driving. In 2022 International Conference on Connected Vehicle and Expo (ICCVE). 1–6. https://doi.org/10.1109/ICCVE52871. 2022.9742859

  64. [72]

    Wei Ye, Jason Carpenter, Zejun Zhang, Rostand A. K. Fezeu, Feng Qian, and Zhi-Li Zhang. 2023. A Closer Look at Stand-Alone 5G Deploy- ments from the UE Perspective. In 2023 IEEE International Mediter- ranean Conference on Communications and Networking (MeditCom) . IEEE, 86–91

  65. [73]

    Wei Ye, Xinyue Hu, Steven Sleder, Anlan Zhang, Udhaya Kumar Day- alan, Ahmad Hassan, Rostand A. K. Fezeu, Akshay Jajoo, Myungjin Lee, Eman Ramadan, Feng Qian, and Zhi-Li Zhang. 2024. Dissecting Carrier Aggregation in 5G Networks: Measurement, QoE Implications and Pre- diction....

  66. [74]

    Motor Trend. 2022. Tech Company Testing Remote Operators as Self- Driving Car Backups. https://www.motortrend.com/news/mira-self- driving-car-remote-control-car/

  67. [78]

    Dongzhu Xu, Anfu Zhou, Xinyu Zhang, Guixian Wang, Xi Liu, Con- gkai An, Yiming Shi, Liang Liu, and Huadong Ma. 2020. Understanding Operational 5G: A First Measurement Study on Its Coverage, Perfor- mance and Energy Consumption. In Proc. of ACM SIGCOMM. 479–494. https://doi.org...

  68. [81]

    Tao Zhang. 2020. Toward Automated Vehicle Teleoperation: Vision, Opportunities, and Challenges. IEEE Internet of Things Journal 7, 12 (2020), 11347–11354. https://doi.org/10.1109/JIOT.2020.3028766 10 APPENDIX 10.1 Ethics This study was carried out by the research team, volunte...

  69. [2020]

    Beyond Throughput, the next Generation: A 5G Dataset with Channel and Context Metrics. InProc. of ACM MMSys. 303–308. https: //doi.org/10.1145/3339825.3394938

  70. [2021]

    In Proceedings of the 2021 ACM SIGCOMM 2021 Conference (Virtual Event, USA) (SIGCOMM ’21)

    A Variegated Look at 5G in the Wild: Performance, Power, and QoE Implications. In Proceedings of the 2021 ACM SIGCOMM 2021 Conference (Virtual Event, USA) (SIGCOMM ’21). Association for Computing Machinery, New York, NY, USA, 610–625. https: //doi.org/10.1145/3452296.3472923

  71. [2022]

    Journal of King Saud University - Computer and Information Sciences 34, 7 (2022), 4135–4162

    A survey on vehicular task offloading: Classification, issues, and challenges. Journal of King Saud University - Computer and Information Sciences 34, 7 (2022), 4135–4162. https://doi.org/10.1016/j.jksuci.2022. 05.016

  72. [2023]

    arXiv:2310.11000 [cs.NI]

    Mid-Band 5G: A Measurement Study in Europe and US. arXiv:2310.11000 [cs.NI]

  73. [6386]

    https://doi.org/10.17487/RFC6386

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.