Pith. sign in

REVIEW 3 major objections 6 minor 30 references

This paper introduces nuTruck, a benchmark that trains and evaluates autonomous-driving planners on distributed electric-drive trucks (DETs), claiming that the conventional split—planner ensures collision avoidance, controller handles rollo

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 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-02 03:58 UTC pith:W5LKQRAI

arxiv 2607.13704 v1 pith:W5LKQRAI submitted 2026-07-15 cs.RO

nuTruck: Benchmarking Autonomous Driving Planning for Distributed Electric-drive Trucks

Jinyu Miao , Pu Zhang , Yifei He , Chengyao Zhang , Kun Jiang , Ke Wang , Mengmeng Yang , Diange Yang This is my paper
classification cs.RO
keywords autonomous driving planningbenchmarkdistributed electric-drive truckrollover safetyclosed-loop simulationreinforcement learning plannerload transfer ratiovehicle dynamics
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.

The paper claims that for distributed electric-drive trucks, collision-free planning is insufficient: a kinematically feasible trajectory can be dynamically unsafe and cause rollover, and no existing benchmark measures this. It builds nuTruck, a closed-loop simulation that combines real-world driving scenarios with a nonlinear six-wheel, three-axle truck model, and adds a Non-Rollover Score to the standard collision-focused metrics. Experiments show that popular trajectory planners (IDM, PDM, Plan-R1) achieve good collision avoidance but very low rollover safety, while an RL action planner that directly outputs wheel torques and steering angles, trained with hand-designed action patterns, attains markedly higher rollover safety with comparable route progress. If the results hold, DET planning must treat dynamic safety as a planning objective, not a controller afterthought.

Core claim

The central claim is that the standard paradigm for autonomous truck planning—letting the planner ensure kinematic passability and letting the low-level controller handle dynamical safety—breaks down for DETs. On the nuTruck benchmark, trajectory-based planners produce paths that, although collision-free, trigger high lateral load transfer and exceed the rollover threshold, yielding very low Non-Rollover Scores. Adding a rollover penalty to the tracking controller improves those scores but degrades tracking accuracy and route completion. In contrast, an RL planner that predicts low-level control signals directly, and uses action priors that exploit DET kinematic structure (e.g., front-rear o

What carries the argument

The load-bearing mechanism is the DET dynamical model integrated into the benchmark: a nonlinear model of a three-axle truck with six independently driven and steered wheels, using a simplified Pacejka tyre formula, rigid-body forces, and dynamic vertical load transfer to compute the LTR, a real-time rollover-risk indicator. The Non-Rollover Score distills LTR traces into a safety metric used in the closed-loop evaluation. The same dynamical model also powers the iLQR tracking controller, so both planning and tracking share the same physics, allowing the benchmark to attribute safety outcomes to the planner rather than to mismatched simulators.

Load-bearing premise

The empirical conclusions rest on the assumption that the simplified rigid-body rollover proxy—a single LTR with constant center-of-gravity height, no suspension articulation, and a hand-chosen threshold of 0.9—faithfully represents the rollover behavior of a real distributed electric-drive truck; if real trucks roll over under different conditions, the reported safety scores and the relative ranking of planners may not transfer to physical vehicles.

What would settle it

Run a high-fidelity multi-body simulation or a physical test with an instrumented DET executing the same collision-free trajectories used in the paper. Measure actual wheel lift-off, lateral load transfer, or suspension loads and compare rollover events against the LTR threshold. If trajectories with LTR below 0.9 cause wheel lift-off, or LTR above 0.9 does not, the benchmark's safety metric is miscalibrated. Also, re-evaluate the baseline planners using a dynamical model validated against real truck measurements; if a trajectory planner with an advanced controller achieves NRS comparable to t

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

If this is right

  • Trajectory planners trained or evaluated only on kinematic benchmarks will be judged unsafe on DETs once rollover dynamics are introduced, so future truck planner benchmarks should include a dynamics-based safety metric like NRS.
  • Hand-designed action priors, such as front-rear opposite steering or Ackermann steering with locked rear wheels, are an effective way to reduce the action space for learning-based truck planners, improving convergence and final route completion.
  • Reinforcement-learning action planners, even without an explicit rollover penalty, achieve high rollover safety because their smooth actions and comfort-motivated rewards produce less aggressive lateral accelerations than trajectory samplers.
  • Adding rollover prevention only in the controller is insufficient; it either fails to eliminate rollover or sacrifices tracking accuracy too much to maintain route progress.
  • The benchmark's metrics, CLS-Safe and NRS, jointly capture collision avoidance and rollover-free progress, offering a template for evaluating planners on other high-risk vehicle types.

Where Pith is reading between the lines

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

  • The paper's qualitative conclusion, that trajectory planners risk rollover while action planners are safer, likely transfers to other heavy vehicles, but the precise threshold and relative gains depend on the rollover model's fidelity; a more detailed suspension model might change which planner wins.
  • A natural next step would be to train an RL planner with an explicit rollover-penalty term in the reward, which the paper does not do; the data suggest this would further narrow the gap between rollover safety and route completion.
  • The benchmark framework could be adapted to other distributed-drive platforms, such as multi-axle military vehicles or vehicles with higher centers of gravity, making rollover-aware evaluation a standard layer for any high-risk autonomous vehicle.
  • The action-pattern results imply that the full 12-dimensional control space is unnecessarily hard for learning; a curriculum that starts with low-dimensional priors and progressively releases dimensions could unlock the full DET capability while retaining training stability.

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

3 major / 6 minor

Summary. The paper introduces nuTruck, a closed-loop simulation benchmark for autonomous driving planning on distributed electric-drive trucks (DETs). It builds on nuPlan scenarios, adds a six-wheel nonlinear truck model with independent drive/steer, defines a lateral-load-transfer-based rollover metric (NRS), and evaluates several rule-based and learning-based planners under closed-loop simulation. The central claim is that conventional trajectory planners combined with a tracking controller—even one augmented with rollover prevention—fail to balance kinematic passability and dynamical safety, whereas RL-based action planners trained in the same simulator achieve higher rollover safety. The paper argues that DET planning must incorporate dynamical safety directly.

Significance. If the results hold, nuTruck would be a useful contribution as the first closed-loop, dynamics-aware benchmark specifically for DET planning. The use of real-world nuPlan scenarios, the six-wheel distributed-drive/steer model, and the explicit rollover metric are genuine strengths. The model is validated against TruckSim with low reported errors, and the baseline comparison across rule-based, IL+RL, and action planners is broad. However, a load-bearing flaw in the iLQR-S cost function undermines the paper's headline experimental evidence: as written, the cost drives LTR toward the rollover threshold rather than away from it. Because the conclusion that the conventional planner/controller separation 'falls short' rests on the iLQR-S comparison, the paper needs a substantive correction before the claims can be accepted.

major comments (3)
  1. [Sec. III-D, Eq. (12)] The iLQR-S cost l' = l + 0.5 Q_l LTR^2 + 0.5 Q_v (LTR - s_max)^2 does not reduce rollover risk. Since LTR in Eq. (8) is an absolute value and hence nonnegative, the two LTR-dependent quadratic terms are minimized at LTR* = Q_v s_max/(Q_l + Q_v), which equals s_max = 0.9 when Q_l = 0 and is strictly positive for any finite Q_l. The derivative at LTR = 0 is -Q_v s_max < 0, so the controller actively pushes LTR away from zero toward the rollover threshold. Thus iLQR-S is not a rollover-prevention controller as claimed in Sec. III-D and Sec. IV-C. Consequently, the 'iLQR+S' rows of Table I and the iLQR-S columns of Table V cannot support the conclusion that 'the conventional paradigm, in which the planner ensures collision avoidance and the controller handles rollover prevention, falls short.' Please replace the penalty with a formulation that actually penalizes large LTR (e.g., a hinge/barr
  2. [Sec. III-B-4, Eq. (7)] The text states that the overall vertical load at each wheel is F_z,i = F_s_z,i + F_d,lon_z,i + F_d,lat_z,i, but Eq. (7) only provides aggregate sums of longitudinal and lateral load transfer over all wheels. No rule is given to distribute these transfers across the three axles or between left and right wheels. Since F_z,i is used in the rolling-resistance model and in the LTR computation, the model is underdetermined and not reproducible as written. Please specify the per-wheel distribution (e.g., based on axle positions, suspension geometry, or the TruckSim configuration).
  3. [Sec. IV-B and III-D] The dynamical model is validated only against TruckSim software, and the iLQR controller uses the same model for its rollout as the simulator itself. This perfect-model assumption likely makes the tracking results in Tables I and V optimistic compared to a real deployment where model mismatch exists. The manuscript should qualify the 'high-fidelity' and 'realistic' claims accordingly, and ideally include a robustness experiment with model parameter perturbations or a comparison against a model different from the one used by the controller. This is not fatal to the benchmark concept, but it is important for interpreting the quantitative NRS scores and the transferability of the conclusions to physical DETs.
minor comments (6)
  1. [Sec. III-B-1] Typos: 'Theotically' should be 'Theoretically'; 'paragraphies' should be 'paragraphs'.
  2. [Sec. III-B-2] Eq. (2) is not fully typeset (e.g., 'Bˆαi'), and the text should clarify that the Pacejka parameters B, C, D, E, S_h, S_v are load-independent; the model does not include normal-load dependence in cornering stiffness, which is a limitation worth stating explicitly.
  3. [Sec. III-E] The NRS metric is described only verbally: 'the score decays quadratically from one to zero as a function of the mean LTR over the scenario.' An explicit formula is needed for reproducibility.
  4. [Sec. III-E / Table III] Table III has inconsistent spacing (e.g., '1.051e 4 kg' and '6.609e4 kg·m2'). Please use uniform scientific notation.
  5. [Sec. III-C / IV-A] Training details for the CaRL planners (number of scenarios, environment steps, reward weights, seeds) are not fully provided; the paper should include enough detail for others to reproduce the RL training in the new benchmark.
  6. [Fig. 6] The convergence curves in Fig. 6 appear to show single-run or aggregate trajectories without error bars. Please report variance across seeds or explain the smoothing procedure.

Circularity Check

0 steps flagged

No significant circularity: the benchmark's load-bearing results rest on external validation and external scenario data, not on equations that reduce to their own outputs.

full rationale

The paper is primarily an empirical benchmark contribution rather than a derivation of a prediction from first principles. Its central claims — that nuTruck is the first closed-loop DET planning benchmark and that trajectory planners induce rollover risk — are grounded in external anchors: the vehicle model is validated against TruckSim (Sec. IV-B, Tab. IV), the scenarios are taken from the nuPlan real-world dataset, and the planner baselines (IDM, PDM, Plan-R1) are existing external methods evaluated without DET-specific fine-tuning. The iLQR controllers do reuse the same dynamical model as the simulator (Sec. III-D), which is an oracle-modeling choice that could affect tracking realism, but this is not a circular reduction: the rollover metric is computed from the propagated vehicle state, and the controller's LTR penalty does not by construction force the paper's main comparative conclusions. The CaRL action planners are trained and evaluated inside the same simulator, which limits generalization claims but does not make the evaluation equivalent to its inputs. The potential issue with Eq. (12), where the LTR penalty is minimized at a nonzero LTR, is a possible confound for interpreting the iLQR-S results as evidence about controller-only rollover prevention, but it is a design/correctness concern, not a case of a result being identical by definition to a fitted parameter or cited assumption. No self-citation is load-bearing, and no prediction is defined in terms of the quantity it claims to predict.

Axiom & Free-Parameter Ledger

4 free parameters · 5 axioms · 0 invented entities

No new physical entities are postulated: the FRO/RWL action patterns (Eq. 9) and the NRS/CLS-Safe metrics are design or scoring choices, not entities. The free parameters that the central results depend on are the tire model coefficients (fitted to TruckSim), the rollover thresholds (hand-chosen), and the unreported controller weights. No independent falsifiable handle is attached to any invented quantity because none is invented.

free parameters (4)
  • Pacejka tire parameters B, C, D, E, Sh, Sv = B=6.876, C=1.600, D=12712.300 N, E=-0.2998, Sh=-0.00001 rad, Sv=1.400 N (Table III)
    Calibrated to match the selected TruckSim 2019 reference truck (Sec. IV-A). These values set the lateral-force and LTR behavior; they are matched to an external simulator rather than derived.
  • Rolling resistance coefficient C_r = 0.004 (Table III)
    Hand-chosen constant in the tire model; linearly scales the longitudinal force at each wheel.
  • Rollover thresholds s_max and s_safe = s_max=0.900, s_safe=0.500
    Hand-chosen LTR thresholds that define rollover events and the NRS metric (Sec. III-B.4, III-E). All NRS and CLS-Safe numbers depend on them, and no sensitivity analysis is given.
  • iLQR cost weights Q, R, Q_l, Q_v, terminal factor lambda, horizon N = N=5, lambda=5; Q, R, Q_l, Q_v values not reported
    Controller hyperparameters (Sec. III-D). Q_l and Q_v determine the iLQR-S rollover penalty; their values are unreported, so the safety/accuracy trade-off in Table V cannot be reproduced or audited.
axioms (5)
  • standard math Standard rigid-body vehicle dynamics and Pacejka magic-formula tire model (Eqs. 2-4)
    Accepted vehicle-dynamics modeling framework, cited to [27]; taken as background without derivation.
  • domain assumption Rollover is adequately proxied by LTR computed from instantaneous rigid load transfer (Eqs. 7-8)
    The load-transfer formula uses constant CoG height h and track width w, ignores suspension articulation, and does not specify the per-wheel distribution of the transferred load. This proxy defines rollover events and the NRS metric.
  • domain assumption TruckSim 2019 is an adequate ground truth for real DET rollover dynamics
    Model fidelity and LTR estimates are validated only against TruckSim (Sec. IV-B), never against a physical DET or real telemetry.
  • standard math Euler integration of the ODE (Eq. 1) is sufficiently accurate at the stated 10/100 Hz rates
    State propagation method chosen for efficiency; stability and discretization error are not analyzed.
  • domain assumption nuPlan car-driving scenarios remain representative of DET operation after kinematic-feasibility filtering
    All evaluation scenarios come from nuPlan car logs, filtered by whether a DET can track the expert trajectory without collision (Sec. III-E). This biases the benchmark toward truck-passable, less crowded roads, and the fraction dropped is unreported.

pith-pipeline@v1.3.0-alltime-deepseek · 11620 in / 19754 out tokens · 199424 ms · 2026-08-02T03:58:50.123373+00:00 · methodology

0 comments
read the original abstract

The dominance of traditional rule-based methods in autonomous driving has gradually been replaced by learning-based approaches. While learning-based planners have achieved considerable success in passenger vehicles, their performance on heavy-duty trucks, particularly modern distributed electric-drive trucks (DETs), remains largely unexplored. To facilitate research and application of learning-based planners in DETs, this letter presents the first high-fidelity benchmark, called nuTruck, designed to support large-scale neural network training and closed-loop evaluation. Given the complex dynamics and high rollover susceptibility of DETs, we first incorporate a highly accurate nonlinear truck dynamical model into the simulation, which enables independent driving and steering of all wheels and captures dynamic load transfer caused by acceleration, deceleration, and cornering, thereby allowing quantitative assessment of rollover risk in closed-loop simulation. Second, we adapt several rule-based and learning-based planners as baselines for DETs and evaluate their performance in closed-loop simulation. Finally, using real-world driving scenarios from the nuPlan dataset, we conduct extensive closed-loop evaluations, analyzing not only conventional collision-free planning performance, but also the dynamical safety of the planned trajectories. The proposed nuTruck benchmark is expected to serve as a new standard for fair and realistic evaluation of autonomous driving planners on DETs.

Figures

Figures reproduced from arXiv: 2607.13704 by Chengyao Zhang, Diange Yang, Jinyu MIao, Ke Wang, Kun Jiang, Mengmeng Yang, Pu Zhang, Yifei He.

Figure 1
Figure 1. Figure 1: A closed-loop trajectory tracking case. Although the controller [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overview framework of the nuTruck benchmark. [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: The action patterns used in CaRL-A2 and CaRL-A4 planners. [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The spatial structure of the implemented DET. [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: A 15-second case study comparing the dynamical state prediction [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: The comparison between CaRL planners [19] trained under different [PITH_FULL_IMAGE:figures/full_fig_p007_6.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

30 extracted references · 2 linked inside Pith

  1. [1]

    Baidu apollo team (2017), apollo: Open source autonomous driving,

    “Baidu apollo team (2017), apollo: Open source autonomous driving,” https://github.com/ApolloAuto/apollo, accessed: 2019-02-11

    2017

  2. [2]

    Robocar: A rapidly deploy- able open source platform for autonomous driving research,

    M. Testouri, G. Elghazaly, and R. Frank, “Robocar: A rapidly deploy- able open source platform for autonomous driving research,”IEEE Intelligent Transportation Systems Magazine, vol. 17, no. 4, pp. 83– 95, 2025

    2025

  3. [3]

    Planning-oriented autonomous driving,

    Y . Hu, J. Yang, L. Chen, K. Li, C. Sima, X. Zhu, S. Chai, S. Du, T. Lin, W. Wang, L. Lu, X. Jia, Q. Liu, J. Dai, Y . Qiao, and H. Li, “Planning-oriented autonomous driving,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2023

    2023

  4. [4]

    Sparsedrive: End-to-end autonomous driving via sparse scene representation,

    W. Sun, X. Lin, Y . Shi, C. Zhang, H. Wu, and S. Zheng, “Sparsedrive: End-to-end autonomous driving via sparse scene representation,”Pro- ceedings of 2025 IEEE International Conference on Robotics and Automation (ICRA), pp. 8795–8801, 2024

    2025

  5. [5]

    nuscenes: A multimodal dataset for autonomous driving,

    H. Caesar, V . Bankiti, A. H. Lang, S. V ora, V . E. Liong, Q. Xu, A. Krishnan, Y . Pan, G. Baldan, and O. Beijbom, “nuscenes: A multimodal dataset for autonomous driving,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020

    2020

  6. [6]

    Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset,

    S. Ettinger, S. Cheng, B. Caine, C. Liu, H. Zhao, S. Pradhan, Y . Chai, B. Sapp, C. R. Qi, Y . Zhou, Z. Yang, A. Chouard, P. Sun, J. Ngiam, V . Vasudevan, A. McCauley, J. Shlens, and D. Anguelov, “Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset,” inProceedings of the IEEE/CVF International Conference on Com...

    2021

  7. [7]

    Nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles,

    C. Holger, K. Juraj, S. T. Kok, F. Whye Kit, W. Eric, L. Alex, F. Luke, B. Oscar, and O. Sammy, “Nuplan: A closed-loop ml-based planning benchmark for autonomous vehicles,” inProceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition workshop (CVPRW), 2021

    2021

  8. [8]

    Re- thinking imitation-based planner for autonomous driving,

    J. Cheng, Y . Chen, X. Mei, B. Yang, B. Li, and M. Liu, “Re- thinking imitation-based planner for autonomous driving,”ArXiv, vol. abs/2309.10443, 2023

    Pith/arXiv arXiv 2023

  9. [9]

    CARLA: An open urban driving simulator,

    A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “CARLA: An open urban driving simulator,” inProceedings of the 1st Annual Conference on Robot Learning (CoRL), 2017, pp. 1–16

    2017

  10. [10]

    Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research,

    C. Gulino, J. Fu, W. Luo, G. Tucker, E. Bronstein, Y . Lu, J. Harb, X. Pan, Y . Wang, X. Chen, J. D. Co-Reyes, R. Agarwal, R. Roelofs, Y . Lu, N. Montali, P. Mougin, Z. Yang, B. White, A. Faust, R. McAllis- ter, D. Anguelov, and B. Sapp, “Waymax: An accelerated, data-driven simulator for large-scale autonomous driving research,” inProceedings of the Advan...

    2023

  11. [11]

    Bench2drive: Towards multi-ability benchmarking of closed-loop end-to-end autonomous driving,

    X. Jia, Z. Yang, Q. Li, Z. Zhang, and J. Yan, “Bench2drive: Towards multi-ability benchmarking of closed-loop end-to-end autonomous driving,” inProceedings of the Advances in Neural Information Processing Systems (NeurIPS), 2024

    2024

  12. [12]

    End-to-end urban driving by imitating a reinforcement learning coach,

    Z. Zhang, A. Liniger, D. Dai, F. Yu, and L. Van Gool, “End-to-end urban driving by imitating a reinforcement learning coach,” inPro- ceedings of 2021 IEEE/CVF International Conference on Computer Vision (ICCV), 2021, pp. 15 202–15 212

    2021

  13. [13]

    Think2drive: Efficient reinforce- ment learning by thinking in latent world model for quasi-realistic autonomous driving (in carla-v2),

    Q. Li, X. Jia, S. Wang, and J. Yan, “Think2drive: Efficient reinforce- ment learning by thinking in latent world model for quasi-realistic autonomous driving (in carla-v2),” inProceedings of the European Conference on Computer Vision (ECCV), 2024

    2024

  14. [14]

    Man truckscenes: A multimodal dataset for autonomous trucking in diverse conditions,

    F. Fent, F. Kuttenreich, F. Ruch, F. Rizwin, S. Juergens, L. Lecher- mann, C. Nissler, A. Perl, U. V oll, M. Yan, and M. Lienkamp, “Man truckscenes: A multimodal dataset for autonomous trucking in diverse conditions,” inProceedings of the Advances in Neural Information Processing Systems (NeurIPS), vol. 37, 2024, pp. 62 062–62 082

    2024

  15. [15]

    Truckdrive: Long-range autonomous highway driving dataset,

    F. Ghilotti, E. Palladin, S. Brucker, A. Sigal, M. Bijelic, and F. Heide, “Truckdrive: Long-range autonomous highway driving dataset,”ArXiv, vol. abs/2603.02413, 2026

    arXiv 2026

  16. [16]

    Congested traffic states in empirical observations and microscopic simulations,

    M. Treiber, A. Hennecke, and D. Helbing, “Congested traffic states in empirical observations and microscopic simulations,”Physical review. E, Statistical physics, plasmas, fluids, and related interdisciplinary topics, vol. 62 2 Pt A, pp. 1805–24, 2000

    2000

  17. [17]

    End-to-end autonomous driving: Challenges and frontiers,

    L. Chen, P. Wu, K. Chitta, B. Jaeger, A. Geiger, and H. Li, “End-to-end autonomous driving: Challenges and frontiers,”IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 46, pp. 10 164–10 183, 2023

    2023

  18. [18]

    Data scaling laws for end-to-end autonomous driving,

    A. Naumann, X. Gu, T. Dimlioglu, M. Bojarski, A. De ˘girmenci, A. Popov, D. Bisla, M. Pavone, U. Muller, and B. Ivanovic, “Data scaling laws for end-to-end autonomous driving,”Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops (CVPRW), pp. 2562–2573, 2025

    2025

  19. [19]

    Carl: Learning scalable planning policies with simple rewards,

    B. Jaeger, D. Dauner, J. Beißwenger, S. Gerstenecker, K. Chitta, and A. Geiger, “Carl: Learning scalable planning policies with simple rewards,” inProceedings of the the Conference on Robot Learning (CoRL), 2025

    2025

  20. [20]

    Learning to drive in a day,

    A. Kendall, J. Hawke, D. Janz, P. Mazur, D. Reda, J.-M. Allen, V .- D. Lam, A. Bewley, and A. Shah, “Learning to drive in a day,” inProceedings of the International Conference on Robotics and Automation (ICRA), 2019, pp. 8248–8254

    2019

  21. [21]

    Urban driver: Learning to drive from real-world demonstrations using policy gradients,

    O. Scheel, L. Bergamini, M. Wolczyk, B. Osi ´nski, and P. Ondruska, “Urban driver: Learning to drive from real-world demonstrations using policy gradients,” inProceedings of the 5th Conference on Robot Learning (CoRL), vol. 164, 08–11 Nov 2022, pp. 718–728

    2022

  22. [22]

    Reward (mis)design for autonomous driving,

    W. B. Knox, A. Allievi, H. Banzhaf, F. Schmitt, and P. Stone, “Reward (mis)design for autonomous driving,”Artificial Intelligence, vol. 316, p. 103829, 2023

    2023

  23. [23]

    An intelligent self-driving truck system for highway transportation,

    D. Wang, L. Gao, Z. Lan, W. Li, J. Ren, J. Zhang, P. Zhang, P. Zhou, S. Wang, J. Pan, D. Manocha, and R. Yang, “An intelligent self-driving truck system for highway transportation,”Frontiers in Neurorobotics, vol. V olume 16 - 2022, 2022

    2022

  24. [24]

    Parting with misconceptions about learning-based vehicle motion planning,

    D. Dauner, M. Hallgarten, A. Geiger, and K. Chitta, “Parting with misconceptions about learning-based vehicle motion planning,” in Proceedings of the Conference on Robot Learning (CoRL), 2023

    2023

  25. [25]

    Plan-r1: Safe and feasible trajectory planning as language modeling,

    X. Tang, M. Kan, S. Shan, and X. Chen, “Plan-r1: Safe and feasible trajectory planning as language modeling,” inThe Fourteenth Interna- tional Conference on Learning Representations, 2026

    2026

  26. [26]

    Constrained iterative lqr for on- road autonomous driving motion planning,

    J. Chen, W. Zhan, and M. Tomizuka, “Constrained iterative lqr for on- road autonomous driving motion planning,” inProceedings of the 2017 IEEE 20th International Conference on Intelligent Transportation Systems (ITSC), 2017, pp. 1–7

    2017

  27. [27]

    The magic formula tyre model,

    H. B. Pacejka and E. Bakker, “The magic formula tyre model,”Vehicle System Dynamics, vol. 21, no. sup001, pp. 1–18, 1992

    1992

  28. [28]

    Pluto: Pushing the limit of imitation learning-based planning for autonomous driving,

    J. Cheng, Y . Chen, and Q. Chen, “Pluto: Pushing the limit of imitation learning-based planning for autonomous driving,”ArXiv, vol. abs/2404.14327, 2024

    Pith/arXiv arXiv 2024

  29. [29]

    Flexible spacing adaptive cruise control using stochastic model predictive control,

    D. Moser, R. Schmied, H. Waschl, and L. del Re, “Flexible spacing adaptive cruise control using stochastic model predictive control,” IEEE Transactions on Control Systems Technology, vol. 26, no. 1, pp. 114–127, 2018

    2018

  30. [30]

    Autonomous driving motion planning with constrained iterative lqr,

    J. Chen, W. Zhan, and M. Tomizuka, “Autonomous driving motion planning with constrained iterative lqr,”IEEE Transactions on Intelli- gent Vehicles, vol. 4, no. 2, pp. 244–254, 2019

    2019