Pith. sign in

REVIEW 4 major objections 6 minor 28 references

NeoRacer is an open, pre-assembled 1:12-scale autonomous race car that delivers research-grade onboard compute and full sensing at about $2,700 so institutions can compare racing algorithms on identical hardware.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

T0 review · grok-4.5

2026-07-30 19:05 UTC pith:DZUQ5MLC

load-bearing objection Solid systems paper: a real pre-assembled Orin Nano 1:12 racer at ~$2.7k with careful architecture and honest openness limits; the benchmarking-standard claim is aspirational, not yet evidenced. the 4 major comments →

arxiv 2607.26855 v1 pith:DZUQ5MLC submitted 2026-07-29 cs.RO cs.SYeess.SY

NeoRacer: An Open, Standardized 1:12 Scale Autonomous Race Car for Benchmarking and Education

classification cs.RO cs.SYeess.SY
keywords autonomous racingrobotics educationopen hardwareROS2SLAMbenchmarkingAckermann steeringsim-to-real
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Autonomous systems research still lacks a widely shared open hardware baseline, so racing and control results are hard to compare and expensive to teach. This paper presents NeoRacer, a standardized Ackermann-steered 1:12 platform that ships pre-assembled with a 67-TOPS Jetson Orin Nano, 270° LiDAR, 120 fps global-shutter camera, and 9-axis IMU for roughly $2,700—more than three times the compute of common educational racers and less than half the price of the nearest pre-assembled research alternative. Modular hardware, a real-time embedded controller board, preconfigured ROS2 drivers, a stable student-facing library, and a matched in-browser simulator are meant to let code move from sim to car without rewrite. Pilot classroom use drove concrete revisions (power board and camera frame rate). The authors argue that spec-locked identical units, open licenses, and repeatable manufacturing can turn autonomous racing into a reproducible benchmark and a broader education pathway.

Core claim

The paper’s central claim is that a single open, manufacturable, pre-assembled 1:12 Ackermann racer can close the gap between underpowered classroom kits and multi-thousand-dollar research builds by locking a research-grade compute-and-sensor suite at institutional-friendly cost, thereby making cross-lab algorithm differences attributable to software rather than hardware.

What carries the argument

Spec-locked NeoRacer platform: identical shipped hardware plus a two-computer split (Orin Nano for autonomy; ESP32-based OSCORE board for real-time actuation, IMU, encoder, and safety), unified ROS2 drivers, a stable high-level library API, and a simulator that exposes the same API so sim code runs unchanged on the car.

Load-bearing premise

That shipping identical units with shared drivers and a matched simulator is enough to create a real cross-institution benchmarking standard and competition ecosystem, even though multi-lab algorithm comparisons, a finished curriculum, and the league structure are still prospective and full board fabrication files are not yet public.

What would settle it

Have several independent groups flash the same public software release on separately purchased units, run the same published racing stack on a common track specification, and check whether lap-time and tracking gaps collapse to algorithm differences; large residual spreads from batch drift, incomplete openness, or non-adoption would falsify the standardization claim.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

Share X Bluesky LinkedIn Reddit HN

If this is right

  • Labs can quote hardware/software/manufacturing identifiers and compare SLAM, planning, and control results without reconciling custom builds.
  • Classroom programs can start from unboxing rather than multi-vendor assembly and fabrication.
  • Sim-to-real learning and certified neural control methods can be re-tested on a shared ground-vehicle substrate.
  • A spec-locked grid enables competition formats where ranking tracks software innovation, not sensor or compute upgrades.
  • Volume manufacturing and kit sponsorship can extend research-grade AV hardware beyond well-funded labs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Until PCB fabrication sources ship, “open hardware” mainly supports repair and interfacing, so third-party clones and true community forks remain constrained.
  • The real adoption test is whether outside groups publish NeoRacer baselines and leaderboards, not only whether pilots succeed at the originating sites.
  • A single high-TOPS onboard budget makes resource-contention and real-time co-scheduling research newly measurable in closed-loop lap metrics rather than microbenchmarks alone.
  • If the matched simulator API stays faithful, virtual qualifying could become the default low-cost entry before any physical purchase.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The manuscript introduces NeoRacer, a pre-assembled 1:12 Ackermann autonomous racing platform built around a Jetson Orin Nano (67 TOPS), 270° LiDAR, 120 fps global-shutter camera, 9-axis IMU, and a custom ESP32-S3 real-time board (OSCORE). It positions the platform between underpowered classroom kits and costly research racers, claiming ~$2,700 retail, over 3× the compute of F1Tenth-class educational platforms, and less than half the price of the nearest pre-assembled alternative, with CERN-OHL-S v2 / GPLv3 licensing, a three-identifier release/traceability scheme, ROS2 drivers, a matched in-browser simulator API, and design lessons from pilot use (MIT IAP emphasized; BU mentioned). The paper is primarily an artifact and infrastructure description: architecture (§3), manufacturing identity (§4), software (§5), intended research uses (§6), a reference curriculum outline (§7), cost framing (§8), pilot revisions (§9), and a planned competition pathway (§10).

Significance. If the cost, compute, pre-assembly, and production-repeatability claims hold, NeoRacer is a useful shared substrate for AV education and small-scale racing research: identical units plus a stable API and release identity would reduce hardware confounds that currently block cross-lab comparison. Strengths include a concrete system architecture (Fig. 2), an explicit safety stack, a driver/topic table (Table 2), honest partial-openness language in §12 about missing Gerbers, and pilot-driven hardware changes (power board; 30→120 fps camera) that are specific and falsifiable. The contribution is infrastructure rather than a new algorithm; its value depends on accurate scoping of what is shipped today versus what is aspirational (curriculum, competition, multi-lab benchmarks).

major comments (4)
  1. [Abstract; §12; §1] Abstract vs §12 openness claim: the abstract states that hardware is open under CERN-OHL-S v2 with “all design files, firmware, and ROS2 packages publicly accessible,” but §12 explicitly withholds PCB fabrication files (Gerbers) and editable EDA sources and limits the current release to documentation, schematics, interfaces, and mechanical models. This is load-bearing for the “open hardware / reproducible shared platform” claim. Align the abstract, introduction bullet list, and conclusion with §12’s actual release boundary, or release the fabrication sources before claiming full design-file accessibility.
  2. [Abstract; §1; §9; Table 3] Two pilots are advertised as informing design (Abstract; §1), including “BU CPS Lab, 10 students,” yet §9 reports only the MIT IAP deployment in substance (revisions in Table 3). The BU pilot is not described (protocol, outcomes, failures, or revisions). Either add a comparable BU subsection with concrete findings or remove/qualify the second-pilot claim so the evidence matches the contribution narrative.
  3. [Abstract; §5.6; §7; §10] §5.6 and the abstract present NeoRacer as already providing “a standardized benchmarking environment” for cross-institutional algorithm comparison. The manuscript shows identical intended specs, drivers, and a release-identity scheme (§4), but no multi-lab or even multi-team algorithm benchmark, no shared task/metrics/logs, and no external adoption data. §7’s curriculum is “in active development” and §10’s competition framework is “still in the ideation phase.” Recast benchmarking/competition language as enabled future use, and state what minimum artifacts (task definitions, datasets, leaderboard protocol, reference baselines) would make the claim operational.
  4. [Table 1; §2.3; §8.1] Table 1 / §2 cost–capability comparison is central to the “3× compute, <½ pre-assembled cost” claim, but several cells mix unlike quantities (DIY BOM vs pre-assembled retail; community vs maintained product; missing AI-accel entries) and F1Tenth is represented primarily by the $6,000 RACECAR/J figure while the official parts BOM (~$3,800) is footnoted. For a fair load-bearing comparison, add a normalized column set (pre-assembled price, parts-only price, assembly skill/tools required, steering type, TOPS, LiDAR range/rate) and ensure every numeric claim in the abstract is traceable to one table row without mixing kit and turnkey prices.
minor comments (6)
  1. [Figure 1; Abstract; §8.1] Figure 1 caption cites a pre-order price of $2,500 while the abstract uses $2,699 and §8 lists $2,700 / $2,500 / $2,200. Harmonize price figures and state the date/terms of each.
  2. [§3.2; Table 2] §3.2 says the camera captures 1920×1200 at 90 fps or 1280×720 at 120 fps; elsewhere “120 fps global shutter” is used generically. Specify which mode is default in racing software and what the driver publishes.
  3. [§2] Related-work coverage of F1Tenth/MuSHR/RACECAR is appropriate; a brief note on other Orin-based 1:10/1:12 efforts or commercial AI racers (beyond DeepRacer’s discontinuation) would help readers judge novelty of the pre-assembled niche.
  4. [§6] §6 reads as a forward-looking brochure for the authors’ prior control/systems papers. Tighten to platform-matched evaluation affordances (sensors, timing path, sim-to-real API) and move lab-specific citations to a shorter paragraph.
  5. [Abstract; §2] Minor typos/wording: “Additonally” (§2.1), “untranslateable” (Abstract), “RACECAR neo” capitalization inconsistencies, and “tenth” vs “Tenth” in places.
  6. [§4; §13] GitHub link is given in the conclusion; add a persistent archival snapshot (Zenodo/Software Heritage) and tag the exact nrlib / HW revision described in the paper to match §4’s reproducibility story.

Circularity Check

0 steps flagged

No significant circularity: descriptive hardware/systems paper with external cost benchmarks and non-load-bearing self-citations.

full rationale

NeoRacer is an artifact/infrastructure paper, not a fitted-theory or first-principles derivation. Its central claims (compute, sensors, pre-assembled price, modular architecture, pilot-informed revisions) are descriptive engineering statements supported by architecture (§3), external BOM/vendor price comparisons (Table 1, §2, §8), release identity (§4), drivers/API (§5), and the MIT IAP pilot (§9). Cost and compute multipliers (e.g., 67 TOPS vs Xavier NX 21 TOPS; $2,700 vs $6,000 RACECAR/J) are arithmetic comparisons to third-party platforms, not quantities defined from the authors' own fit targets. Section 6 cites prior work by overlapping authors (RE+AL, CAPS, Omnivisor, MemPol, etc.) only as proposed future research directions the platform could host; the paper explicitly states those directions are not its focus and does not use them as premises that force the platform claims. There is no self-definitional loop, no fitted input renamed as prediction, no uniqueness theorem imported to forbid alternatives, and no renaming of a known empirical law. Aspirational framing that identical units will become a cross-institutional benchmark (§5.6, §10) is a scope/overclaim issue, not circular derivation. Honest finding: no circularity.

Axiom & Free-Parameter Ledger

3 free parameters · 5 axioms · 4 invented entities

Load-bearing content is mostly engineering and market assumptions, not mathematical axioms. The central claim rests on component capability claims, price comparability, the idea that hardware homogeneity yields algorithm-comparable benchmarks, and partial openness being enough for community standardization.

free parameters (3)
  • Retail/pre-order/volume price points ($2700/$2500/$2200) = $2700 retail; $2500 pre-order; $2200 at 15+ units
    Chosen commercial prices that anchor the affordability half of the central claim; not derived from a public full BOM cost model in the paper.
  • Target class sizing and course shape (12 teams of 4; 12-week course) = 48 students / 12 teams; 12 weeks
    Curriculum reference design parameters set by authors for packaging and budget examples, not measured optima.
  • Camera racing mode (1280x720 at 120 fps) and speed regime (~6 m/s) = 120 fps; up to ~6 m/s
    Operating points selected after pilot failure of 30 fps; they define the perception adequacy claim.
axioms (5)
  • domain assumption Cross-institution algorithm benchmarking becomes meaningful primarily when sensing, compute, and drivers are hardware-identical.
    Stated throughout abstract, §1, §5.6, and §10 as the justification for spec-locked production; unproven by multi-lab trials in this paper.
  • domain assumption Pre-assembled Orin Nano + LakiBeam1 + 120 fps global-shutter camera + OSCORE is sufficient to run concurrent SLAM, deep learning, and real-time control for racing education/research.
    Asserted in §3.1 and §5.3; plausible from compute ratings but not demonstrated with measured onboard latency/throughput budgets under a fixed autonomy stack.
  • domain assumption Nearest relevant pre-assembled comparator for price is RACECAR/J F1Tenth at $6,000, and DIY F1Tenth parts cost ~$3,800 plus fabrication burden.
    Table 1 and §2 rely on these external price anchors for the 'half the cost' claim.
  • ad hoc to paper CERN-OHL-S documentation without present Gerbers still counts as an open hardware release adequate for community use, repair, and interfacing.
    §12 explicitly withholds fabrication sources while branding the platform open; this boundary condition is author-chosen.
  • domain assumption Pilot classroom stress tests (especially MIT IAP) are adequate validation proxies for reliability and accessibility claims.
    §9 uses one 15-student two-week intensive plus mentioned BU use to justify design maturity.
invented entities (4)
  • OSCORE embedded controller board independent evidence
    purpose: Real-time actuation, IMU/encoder/RC path, USB aggregation of LiDAR Ethernet bridge and peripherals, watchdog safety.
    Custom electronics unique to this platform; central to the two-computer architecture in §3.4.
  • Three-identifier release scheme (NR-HW / nrlib / MFG-...) no independent evidence
    purpose: Make each shipped unit citable and traceable for reproducibility and support.
    Process invention in §4; usefulness depends on community adoption and public matrices not shown here.
  • Neobotics Playground in-browser simulator no independent evidence
    purpose: Same student API as the car for curriculum, virtual qualifying, and sim-to-real transfer.
    Introduced in §5.5/§7/§10 as primary simulation path; no external validation of physical fidelity is reported.
  • racecar-neo-library stable API no independent evidence
    purpose: Insulate curricula and controllers from hardware substitutions across manufacturing runs.
    Software facade derived from MIT RACECAR/BWSI patterns (§5.4); stability claim is prospective.

pith-pipeline@v1.2.0-daily-grok45 · 19721 in / 3787 out tokens · 83628 ms · 2026-07-30T19:05:09.537304+00:00 · methodology

0 comments
Cite this review

Pith. "Pith review of NeoRacer: An Open, Standardized 1:12 Scale Autonomous Race Car for Benchmarking and Education." pith.science (2026). https://pith.science/paper/DZUQ5MLC

@misc{pith2026260726855,
  author       = {Pith},
  title        = {Pith review of: NeoRacer: An Open, Standardized 1:12 Scale Autonomous Race Car for Benchmarking and Education},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DZUQ5MLC}},
  note         = {Machine review of arXiv:2607.26855}
}
Share X Bluesky LinkedIn Reddit HN
read the original abstract

Many scientific fields rely on standard benchmarks and shared platforms to improve review and reproducibility, but autonomous systems research still lacks widely accepted open hardware. Where standardization has emerged, progress has accelerated. This is especially evident in autonomous racing, where teams often build custom systems or buy niche, expensive vehicles, making control and robotics research and education hard to compare and reproduce. High costs also limit access outside well-funded labs, while affordable educational robots are often underpowered. To address this gap, we present NeoRacer, an open-source 1:12 scale autonomous racing platform. It is built around an NVIDIA Jetson Orin Nano (67 TOPS), a 270{\deg} LiDAR, a 120 fps global-shutter camera, and a 9-axis IMU. NeoRacer ships pre-assembled for USD 2,699, offering over 3x the compute of comparable platforms at less than half the cost of the nearest pre-assembled alternative. Co-developed by the Neobotics Foundation and Seeed Studio, and manufactured by Seeed Studio, NeoRacer combines open hardware and software design with scalable, repeatable production. The modular, extensible platform provides a standardized benchmarking environment for autonomous racing algorithms across institutions. We describe the hardware/software architecture, design decisions from two pilot deployments (MIT IAP, 15 students; BU CPS Lab, 10 students), and key cost-performance tradeoffs. Hardware is licensed under CERN-OHL-S v2 and software under GPLv3, with all design files, firmware, and ROS2 packages publicly accessible.

Figures

Figures reproduced from arXiv: 2607.26855 by Ansh Mehta, Bassel El Mabsout, Koneshka Bandyopadhyay, Renato Mancuso.

Figure 1
Figure 1. Figure 1: Operating model for the NeoRacer platform, organized around three activity pillars: platform productization and [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: NeoRacer system architecture. The NVIDIA Jetson Orin Nano hosts perception, planning, and control under [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The NeoRacer autonomous racing platform. Visible subsystems include the Richbeam LakiBeam1 LiDAR (top [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Side profile showing the LiDAR (cylindrical unit, top), camera module (left), independent coilover suspension, and [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Top-down CAD view of the 6061 aluminum alloy chassis, showing the independent suspension geometry at all four [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The Neobotics Playground in-browser simulator. The interface combines a controller-code editor (left), live sensor [PITH_FULL_IMAGE:figures/full_fig_p012_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Matched physical and simulated views of the NeoRacer. The same vehicle model, sensor placements, and dimen [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

28 extracted references · 2 canonical work pages

  1. [1]

    Paull, J

    L. Paull, J. Tani, H. Ahn, J. Alonso-Mora, L. Carlone, M. Cap, Y . F. Chen, C. Choi, J. Dusek, Y . Fang, et al., Duckietown: An open, inexpensive and flexible platform for autonomy education and research, in: IEEE International Conference on Robotics and Automation (ICRA), IEEE, 2017, pp. 1497–1504. 17

  2. [2]

    SunFounder, Picar-x,https://www.sunfounder.com/products/picar-x, accessed: 2026-04-05 (2023)

  3. [3]

    O’Kelly, H

    M. O’Kelly, H. Zheng, D. Karthik, R. Mangharam, F1tenth: An open-source evaluation envi- ronment for continuous control and reinforcement learning, in: NeurIPS 2019 Competition and Demonstration Track, PMLR, 2020, pp. 77–89

  4. [4]

    S. S. Srinivasa, P. Lancaster, J. Michalove, M. Schmittle, C. Summers, M. Rockett, J. R. Smith, S. Choudhury, C. Mavrogiannis, F. Sadeghi, MuSHR: A low-cost, open-source robotic racecar for education and research, arXiv preprint arXiv:1908.08031 (2019)

  5. [5]

    For Inspiration and Recognition of Science and Technology, FIRST robotics competition, https://www.firstinspires.org, accessed: 2026-04-04 (2000)

  6. [6]

    REC Foundation, VEX Robotics Competition,https://www.vexrobotics.com/ competition, accessed: 2026-04-05 (2007)

  7. [7]

    Ac- cessed: 2026-04-05 (2024)

    RACECAR/J, RACECAR/J robot F1 Tenth,https://racecarj.com/collections/ f1-tenth/products/racecar-j-robot-f1-tenth, $6,000 pre-assembled. Ac- cessed: 2026-04-05 (2024)

  8. [8]

    Ac- cessed: 2026-04-05 (2024)

    Acroname, Hokuyo UST-10LX scanning laser rangefinder,https://acroname.com/ store/scanning-laser-rangefinder-ethernet-r359-ust-10lx, $1,200. Ac- cessed: 2026-04-05 (2024)

  9. [9]

    Accessed: 2026-04-05 (2024)

    F1TENTH, F1TENTH bill of materials,https://f1tenth.readthedocs.io/en/ foxy_test/getting_started/build_car/bom.html, total approximately $3,800. Accessed: 2026-04-05 (2024)

  10. [10]

    Accessed: 2026-04-05 (2024)

    Acroname, Hokuyo UTM-30LX-EW scanning laser rangefinder,https://acroname.com/ store/r354-utm-30lx-ew, $4,375. Accessed: 2026-04-05 (2024)

  11. [11]

    Karaman, A

    S. Karaman, A. Anders, M. Boulet, J. Connor, K. Gregson, W. Guerra, O. Guldner, M. Mo- hamoud, B. Plancher, R. Shin, J. Vivilecchia, Project-based, collaborative, algorithmic robotics for high school students: Programming self-driving race cars at MIT, in: 2017 IEEE Inte- grated STEM Education Conference (ISEC), IEEE, 2017.doi:10.1109/ISECon.2017. 7910242

  12. [12]

    github.io/, accessed: 2026-04-05 (2023)

    Beaver Works Summer Institute, RACECAR Neo,https://mit-bwsi-racecar-mn. github.io/, accessed: 2026-04-05 (2023)

  13. [13]

    Accessed: 2026-04-05 (2018)

    Amazon Web Services, AWS DeepRacer,https://aws.amazon.com/deepracer/, ser- vice discontinued December 2025. Accessed: 2026-04-05 (2018)

  14. [14]

    Clearpath Robotics, Turtlebot 4,https://clearpathrobotics.com/turtlebot-4/, accessed: 2026-04-04 (2022)

  15. [15]

    Husarion, ROSbot 3 PRO,https://husarion.com/manuals/rosbot3pro/, accessed: 2026-04-05 (2024)

  16. [16]

    MiRobot, XBot 4WD platform materials,https://www.mirobot.ai/storeitems/p/ xbot, accessed May 2026. 18

  17. [17]

    Robotics, Clearpath Robotics Jackal unmanned ground vehicle materials,https: //clearpathrobotics.com/jackal-small-unmanned-ground-vehicle/, ac- cessed May 2026

    C. Robotics, Clearpath Robotics Jackal unmanned ground vehicle materials,https: //clearpathrobotics.com/jackal-small-unmanned-ground-vehicle/, ac- cessed May 2026

  18. [18]

    Mysore, B

    S. Mysore, B. Mabsout, K. Saenko, R. Mancuso, How to train your quadrotor: A framework for consistently smooth and responsive flight control via reinforcement learning, ACM Transactions on Cyber-Physical Systems 5 (4) (2021).doi:10.1145/3466618

  19. [19]

    Mysore, B

    S. Mysore, B. Mabsout, R. Mancuso, K. Saenko, Regularizing action policies for smooth control with reinforcement learning, in: 2021 IEEE International Conference on Robotics and Automa- tion (ICRA), IEEE, 2021, pp. 1810–1816.doi:10.1109/ICRA48506.2021.9561138

  20. [20]

    El Mabsout, A

    B. El Mabsout, A. Abdelgawad, R. Mancuso, Closing the intent-to-behavior gap via fulfillment priority logic, in: 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), IEEE, 2025, pp. 4174–4181.doi:10.1109/IROS60139.2025.11247623

  21. [21]

    Mysore, B

    S. Mysore, B. Mabsout, R. Mancuso, K. Saenko, Honey, i shrunk the actor: A case study on preserving performance with smaller actors in actor-critic rl, in: 2021 IEEE Conference on Games (CoG), IEEE, 2021.doi:10.1109/CoG52621.2021.9619008

  22. [22]

    El Mabsout, S

    B. El Mabsout, S. Roozkhosh, S. Mysore, K. Saenko, R. Mancuso, Sim-anchored learning for on-the-fly adaptation, arXiv preprint arXiv:2301.06987 (2023)

  23. [23]

    Abdelgawad, B

    A. Abdelgawad, B. El Mabsout, Z. Wang, R. Mancuso, S. B. Andersson, R. Tron, Accelerating lyapunov-stable neural control using fulfillment priority logic, in: Proceedings of the American Control Conference (ACC), New Orleans, LA, USA, 2026

  24. [24]

    Ottaviano, F

    D. Ottaviano, F. Ciraolo, R. Mancuso, M. Cinque, The Omnivisor: A real-time static partition- ing hypervisor extension for heterogeneous core virtualization over MPSoCs, in: 36th Euromi- cro Conference on Real-Time Systems (ECRTS 2024), 2024, pp. 7:1–7:27.doi:10.4230/ LIPIcs.ECRTS.2024.7

  25. [25]

    Sohal, R

    P. Sohal, R. Tabish, U. Drepper, R. Mancuso, E-WarP: A system-wide framework for memory bandwidth profiling and management, in: 2020 IEEE Real-Time Systems Symposium (RTSS), 2020, pp. 345–357.doi:10.1109/RTSS49844.2020.00039

  26. [26]

    Zuepke, A

    A. Zuepke, A. Bastoni, W. Chen, M. Caccamo, R. Mancuso, MemPol: Policing core memory bandwidth from outside of the cores, in: 2023 IEEE 29th Real-Time and Embedded Technol- ogy and Applications Symposium (RTAS), 2023, pp. 235–248.doi:10.1109/RTAS58335. 2023.00026

  27. [27]

    W. Chen, I. Izhbirdeev, R. Mancuso, CAPA: A framework for contention-aware and progress- aware multi-core real-time systems, IEEE Transactions on Computers (2026) 1–14doi:10. 1109/TC.2026.3688446

  28. [28]

    Mancuso, S

    R. Mancuso, S. Roozkhosh, D. Hoornaert, J. H. Mun, T. I. Papon, M. Athanassoulis, Software- shaped platforms, in: Proceedings of Cyber-Physical Systems and IoT Week 2023, 2023.doi: 10.1145/3576914.3587546. 19