pith. sign in

arxiv: 1907.05738 · v1 · pith:ZASKKW5Enew · submitted 2019-07-12 · 💻 cs.CV · cs.RO· cs.SY· eess.SY

Learning a Curve Guardian for Motorcycles

Simon Hecker , Alexander Liniger , Henrik Maurenbrecher , Dengxin Dai , Luc Van Gool This is my paper

Pith reviewed 2026-05-24 22:20 UTC · model grok-4.3

classification 💻 cs.CV cs.ROcs.SYeess.SY
keywords motorcycle safetycurve warningintra-lane positionroll angle estimationconvolutional neural networksroad curvatureoptimal controlmap-based prediction
0
0 comments X

The pith

A system using CNNs for motorcycle lane position and roll angle plus map data predicts safer curve trajectories than prior warning methods.

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

The paper introduces a curve warning system for motorcycles that addresses the fact that up to 17 percent of accidents occur in curves mainly due to wrong speed and intra-lane position. It trains convolutional neural networks to estimate the rider's position within the lane and the bike's roll angle, upgrades an optimal control model to include road incline, and pulls accurate future road geometry from a map database. These components together generate predicted trajectories that the authors show are more accurate and safer in real-world tests. Two new datasets are released to support training and evaluation of such vision-based motorcycle estimators. The result is a scalable warning setup that does not require extensive additional hardware on the vehicle.

Core claim

The paper claims that combining CNN-based estimates of intra-lane position and roll angle with an incline-aware controller model and HERE map road geometry produces trajectory predictions that are both more accurate and safer for warning motorcyclists in curves.

What carries the argument

Two convolutional neural networks that output motorcycle intra-lane position and roll angle, fused with an optimal control model that adds road incline and with map-derived future path geometry.

If this is right

  • Warnings can be issued earlier because the system forecasts the motorcycle's path several seconds ahead using the map.
  • The approach works on any road covered by the map database without needing to instrument every curve with sensors.
  • Releasing the two datasets allows other researchers to train and compare vision models for motorcycle state estimation.
  • The incline term in the controller reduces mismatch between predicted and actual dynamics on sloped roads.

Where Pith is reading between the lines

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

  • The same CNN-plus-map pipeline could be tested on bicycles or electric scooters to see whether the accuracy gains transfer to lighter two-wheelers.
  • Adding real-time traffic or weather overlays to the map layer might further refine the safe-trajectory predictions.
  • If the position and roll networks prove robust, they could serve as input to a future autonomous lane-keeping controller for motorcycles.

Load-bearing premise

The neural networks trained on the collected datasets will continue to estimate position and roll accurately under new lighting, weather, and road conditions, and the map data will stay precise enough for reliable future-path predictions.

What would settle it

Run the full system on a fleet of motorcycles through hundreds of real curves while recording ground-truth position, roll, and actual rider paths with high-precision sensors; the claim fails if predicted safe trajectories deviate enough to miss known high-risk situations or produce false warnings at rates worse than existing systems.

Figures

Figures reproduced from arXiv: 1907.05738 by Alexander Liniger, Dengxin Dai, Henrik Maurenbrecher, Luc Van Gool, Simon Hecker.

Figure 1
Figure 1. Figure 1: Camera rig configuration for (a) Learning Dataset and (b) Motorcycle Dataset. Note that in reality all 7 cameras of (a) record. 1) Learning Dataset: Our networks are trained by either supplying images recorded at different locations within a lane or, in addition, augmented via a rotation to simulate and learn to predict motorcycle roll. It would be difficult to collect large-scale accurate lane location da… view at source ↗
Figure 2
Figure 2. Figure 2: Curvilinear coordinate system We formulate our dynamics in a curvilinear coordinate system, see [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Predictions of LNet compared to the measured ground truth in meters. Sample images shown for locations (A), (B) and (C). 2) RNet: A realistic data test was conducted using a hand￾held GoPro and manually rotating the camera while filming a street in Zurich. Using the GoPro’s built-in IMU, the true roll angle of each taken frame ϕg can be calculated. Note, that the camera was set to be level in terms of pitc… view at source ↗
Figure 4
Figure 4. Figure 4: Predictions of RNet compared to the IMU ground truth in degrees. Sample images shown for locations (A), (B) and (C). C. Case Study For the curve warning system evaluation we present a proof-of-concept case study. We highlight how the changes in the optimal control problem formulation have a significant and positive influence on the system performance. The case study evaluates the system performance on one … view at source ↗
Figure 5
Figure 5. Figure 5: The section of road used in our case study. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Analyzing the influence of the road slope on [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
read the original abstract

Up to 17% of all motorcycle accidents occur when the rider is maneuvering through a curve and the main cause of curve accidents can be attributed to inappropriate speed and wrong intra-lane position of the motorcycle. Existing curve warning systems lack crucial state estimation components and do not scale well. We propose a new type of road curvature warning system for motorcycles, combining the latest advances in computer vision, optimal control and mapping technologies to alleviate these shortcomings. Our contributes are fourfold: 1) we predict the motorcycle's intra-lane position using a convolutional neural network (CNN), 2) we predict the motorcycle roll angle using a CNN, 3) we use an upgraded controller model that incorporates road incline for a more realistic model and prediction, 4) we design a scale-able system by utilizing HERE Technologies map database to obtain the accurate road geometry of the future path. In addition, we present two datasets that are used for training and evaluating of our system respectively, both datasets will be made publicly available. We test our system on a diverse set of real world scenarios and present a detailed case-study. We show that our system is able to predict more accurate and safer curve trajectories, and consequently warn and improve the safety for motorcyclists.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The paper proposes a curve-warning system for motorcycles that combines CNN-based prediction of intra-lane position and roll angle, an incline-augmented optimal controller, and HERE map data for future road geometry. It releases two datasets (one for training, one for evaluation) and evaluates the full pipeline via a detailed case study on real-world riding scenarios, claiming that the system produces more accurate and safer trajectories than existing approaches.

Significance. If the performance claims can be substantiated with quantitative metrics, the work would address a concrete safety problem (curve-related motorcycle accidents) by integrating computer vision, control, and mapping in a scalable way. The public release of the two datasets is a clear positive contribution that could enable follow-on research in motorcycle state estimation.

major comments (2)
  1. [Evaluation section] Evaluation section: The central claim that the system 'predict[s] more accurate and safer curve trajectories' rests on a qualitative case study alone. No quantitative error metrics (e.g., position or roll prediction RMSE), baseline comparisons, ablation results, or sensitivity analysis to CNN noise or map inaccuracies are reported, leaving the safety-margin assertion unsupported.
  2. [§4 and Evaluation] §4 (Controller) and Evaluation: The upgraded incline-augmented controller is presented as a key contribution, yet the case study provides no numerical comparison of predicted trajectories or warning times with versus without the incline term, so the incremental benefit cannot be assessed.
minor comments (2)
  1. [Abstract] Abstract: 'Our contributes are fourfold' should read 'Our contributions are fourfold.'
  2. [Abstract and Datasets section] The manuscript states that both datasets 'will be made publicly available' but does not provide a link, DOI, or repository reference in the current version.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive comments. We address each major point below and will revise the manuscript accordingly to strengthen the evaluation.

read point-by-point responses

  1. Referee: [Evaluation section] Evaluation section: The central claim that the system 'predict[s] more accurate and safer curve trajectories' rests on a qualitative case study alone. No quantitative error metrics (e.g., position or roll prediction RMSE), baseline comparisons, ablation results, or sensitivity analysis to CNN noise or map inaccuracies are reported, leaving the safety-margin assertion unsupported.

    Authors: We agree that the evaluation relies on a qualitative case study and that quantitative metrics are needed to substantiate the claims. In the revised manuscript we will add RMSE values for intra-lane position and roll-angle predictions, baseline comparisons, ablation studies, and sensitivity analysis to CNN noise and map inaccuracies. revision: yes

  2. Referee: [§4 and Evaluation] §4 (Controller) and Evaluation: The upgraded incline-augmented controller is presented as a key contribution, yet the case study provides no numerical comparison of predicted trajectories or warning times with versus without the incline term, so the incremental benefit cannot be assessed.

    Authors: We acknowledge the absence of a direct numerical comparison for the incline term. The revision will include an explicit comparison (with versus without the incline augmentation) of predicted trajectories and warning times to quantify its incremental benefit. revision: yes

Circularity Check

0 steps flagged

No significant circularity; system relies on external data and trained models

full rationale

The paper's core contributions are training separate CNNs on provided external datasets for intra-lane position and roll-angle prediction, integrating HERE map geometry as an external input, and augmenting a controller with incline terms. These steps are standard supervised learning plus external data sources; the case-study evaluation does not reduce any claimed prediction back to a fitted parameter or self-citation by construction. No equations or derivations are shown that equate outputs to inputs via self-definition or renaming.

Axiom & Free-Parameter Ledger

2 free parameters · 2 axioms · 0 invented entities

The system relies on trained neural network parameters and assumptions about map accuracy and model fidelity; no new physical entities invented.

free parameters (2)
  • CNN model weights for position prediction
    Trained on the new dataset, these are fitted parameters central to the position estimation.
  • CNN model weights for roll angle prediction
    Trained on the new dataset for roll estimation.
axioms (2)
  • domain assumption The road geometry from HERE map is accurate for the future path.
    Used to obtain accurate road geometry without verification in the abstract.
  • domain assumption The upgraded controller model with road incline accurately represents motorcycle dynamics.
    Incorporates road incline for more realistic prediction.

pith-pipeline@v0.9.0 · 5764 in / 1370 out tokens · 23063 ms · 2026-05-24T22:20:42.498690+00:00 · methodology

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

29 extracted references · 29 canonical work pages · 1 internal anchor

  1. [1]

    Maids: In-depth investigation of accidents involving powered twowheelers,

    ACEM, “Maids: In-depth investigation of accidents involving powered twowheelers,” 2004

    work page 2004

  2. [2]

    Trace project. deliverable 1.3. road users and accident causation. part 3: Summary report,

    A. Molinero, J. M. Perandones, T. Hermitte, A. Grimaldi, J. Gwe- hengerber, D. Daschner, J. M. Barrios, A. Aparicio, S. Schick, P. Van Elslande, and K. Fouquet, “Trace project. deliverable 1.3. road users and accident causation. part 3: Summary report,” 01 2008

    work page 2008

  3. [3]

    Experimental evaluation of a system for assisting motorcyclists to safely ride road bends,

    F. Biral, P. Bosetti, and R. Lot, “Experimental evaluation of a system for assisting motorcyclists to safely ride road bends,” European Transport Research Review , vol. 6, no. 4, pp. 411–423, 2014

    work page 2014

  4. [4]

    An intelligent curve warning system for powered two wheel vehicles,

    F. Biral, M. Da Lio, R. Lot, and R. Sartori, “An intelligent curve warning system for powered two wheel vehicles,” European transport research review, vol. 2, no. 3, pp. 147–156, 2010

    work page 2010

  5. [5]

    Image-based lateral position, steering behavior estimation, and road curvature prediction for motorcycles,

    P.-M. Damon, H. Hadj-Abdelkader, H. Arioui, and K. Youcef-Toumi, “Image-based lateral position, steering behavior estimation, and road curvature prediction for motorcycles,” IEEE Robotics and Automation Letters, vol. 3, no. 3, pp. 2694–2701, 2018

    work page 2018

  6. [6]

    Roll angle estimation in two- wheeled vehicles,

    I. Boniolo, S. Savaresi, and M. Tanelli, “Roll angle estimation in two- wheeled vehicles,” IET Control Theory & Applications , vol. 3, no. 1, pp. 20–32, 2009

    work page 2009

  7. [7]

    Real-time roll angle estimation for two-wheeled vehicles,

    R. Lot, V . Cossalter, and M. Massaro, “Real-time roll angle estimation for two-wheeled vehicles,” in Biennial Conference on Engineering Systems Design and Analysis . American Society of Mechanical Engineers, 2012

    work page 2012

  8. [8]

    End-to-end learning of driving models with surround-view cameras and route planners,

    S. Hecker, D. Dai, and L. Van Gool, “End-to-end learning of driving models with surround-view cameras and route planners,” in European Conference on Computer Vision (ECCV) , 2018

    work page 2018

  9. [9]

    Learning Accurate, Comfortable and Human-like Driving

    S. Hecker, D. Dai, and L. Van Gool, “Learning accurate, comfortable and human-like driving,” arXiv preprint arXiv:1903.10995 , 2019

    work page internal anchor Pith review Pith/arXiv arXiv 1903

  10. [10]

    Failure prediction for autonomous driving models,

    ——, “Failure prediction for autonomous driving models,” in IEEE Intelligent V ehicles Symposium (IV) , 2018

    work page 2018

  11. [11]

    Inverse Perspective Mapping Roll Angle Estimation for Motorcycles,

    P.-M. Damon, H. Hadj-Abdelkader, H. Arioui, and K. Youcef-Toumi, “Inverse Perspective Mapping Roll Angle Estimation for Motorcycles,” in Conference on Control, Automation, Robotics and Vision (ICARCV) , 2018

    work page 2018

  12. [12]

    Video-based roll angle estimation for two-wheeled vehicles,

    M. Schlipsing, J. Schepanek, and J. Salmen, “Video-based roll angle estimation for two-wheeled vehicles,” in Intelligent V ehicles Sympo- sium (IV) . IEEE, 2011. Fig. 5: The section of road used in our case study. Rnet and LNet predictions for four sample frames are shown, along with the optimal trajectory planned by the controller in red. Fig. 6: Curve w...

    work page 2011

  13. [13]

    Image orientation estimation with convolutional networks,

    P. Fischer, A. Dosovitskiy, and T. Brox, “Image orientation estimation with convolutional networks,” in German Conference on Pattern Recognition. Springer, 2015

    work page 2015

  14. [14]

    On curve negotiation: From driver support to automation,

    P. Bosetti, M. Da Lio, and A. Saroldi, “On curve negotiation: From driver support to automation,” IEEE Transactions on Intelligent Trans- portation Systems , vol. 16, no. 4, pp. 2082–2093, Aug 2015

    work page 2082

  15. [15]

    Parallel autonomy in automated vehicles: Safe motion generation with minimal intervention,

    W. Schwarting, J. Alonso-Mora, L. Pauli, S. Karaman, and D. Rus, “Parallel autonomy in automated vehicles: Safe motion generation with minimal intervention,” in International Conference on Robotics and Automation (ICRA) . IEEE, 2017

    work page 2017

  16. [16]

    Predictive active steering control for autonomous vehicle systems,

    P. Falcone, F. Borrelli, J. Asgari, H. E. Tseng, and D. Hrovat, “Predictive active steering control for autonomous vehicle systems,” IEEE Transactions on control systems technology , vol. 15, no. 3, pp. 566–580, 2007

    work page 2007

  17. [17]

    Optimization-based au- tonomous racing of 1: 43 scale rc cars,

    A. Liniger, A. Domahidi, and M. Morari, “Optimization-based au- tonomous racing of 1: 43 scale rc cars,” Optimal Control Applications and Methods , vol. 36, no. 5, pp. 628–647, 2015

    work page 2015

  18. [18]

    A curvilinear abscissa approach for the lap time optimization of racing vehicles,

    R. Lot and F. Biral, “A curvilinear abscissa approach for the lap time optimization of racing vehicles,” IF AC World Congress, 2014

    work page 2014

  19. [19]

    Model predictive for path following with motorcycles: application to the development of the pilot model for virtual prototyping,

    R. Frezza, A. Beghi, and A. Saccon, “Model predictive for path following with motorcycles: application to the development of the pilot model for virtual prototyping,” in Conference on Decision and Control (CDC), 2004

    work page 2004

  20. [20]

    Motorcycle modeling for high-performance maneuvering,

    J. Hauser and A. Saccon, “Motorcycle modeling for high-performance maneuvering,” IEEE Control Systems Magazine , vol. 26, no. 5, pp. 89–105, Oct 2006

    work page 2006

  21. [21]

    A virtual rider for motorcycles: An approach based on optimal control and maneuver regulation,

    A. Saccon, J. Hauser, and A. Beghi, “A virtual rider for motorcycles: An approach based on optimal control and maneuver regulation,” in Symposium on Communications, Control and Signal Processing . IEEE, 2008

    work page 2008

  22. [22]

    Modelling and model predictive control for a bicycle-rider system,

    T. D. Chu and C. K. Chen, “Modelling and model predictive control for a bicycle-rider system,” V ehicle System Dynamics, vol. 56, no. 1, pp. 128–149, 2018

    work page 2018

  23. [23]

    The viterbi algorithm,

    G. D. Forney, “The viterbi algorithm,” Proceedings of the IEEE , vol. 61, no. 3, pp. 268–278, 1973

    work page 1973

  24. [24]

    Deep residual learning for image recognition,

    K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Conference on Computer Vision and Pattern Recognition (CVPR) , 2016

    work page 2016

  25. [25]

    Forces nlp: an efficient implementation of interior-point methods for multistage nonlinear nonconvex programs,

    A. Zanelli, A. Domahidi, J. Jerez, and M. Morari, “Forces nlp: an efficient implementation of interior-point methods for multistage nonlinear nonconvex programs,” International Journal of Control , 2017

    work page 2017

  26. [26]

    Julia: A fresh approach to numerical computing,

    J. Bezanson, A. Edelman, S. Karpinski, and V . B. Shah, “Julia: A fresh approach to numerical computing,” SIAM review, vol. 59, no. 1, pp. 65–98, 2017

    work page 2017

  27. [27]

    Introduction to ipopt: A tutorial for downloading, installing, and using ipopt,

    A. Waechter, C. Laird, F. Margot, and Y . Kawajir, “Introduction to ipopt: A tutorial for downloading, installing, and using ipopt,” Revision, 2009

    work page 2009

  28. [28]

    Real-time control for autonomous racing based on viability theory,

    A. Liniger and J. Lygeros, “Real-time control for autonomous racing based on viability theory,” IEEE Transactions on Control Systems Technology, vol. 27, no. 2, pp. 464–478, March 2019

    work page 2019

  29. [29]

    Real-time control for at-limit handling driving on a predefined path,

    T. Novi, A. Liniger, R. Capitani, and C. Annicchiarico, “Real-time control for at-limit handling driving on a predefined path,” V ehicle System Dynamics , 2019

    work page 2019