Pith. sign in

REVIEW 3 major objections 8 minor 24 references

Reachability-Based Contingency Planning against Multi-Modal Predictions with Branch MPC

T0 review · 3 major / 8 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper claims that reachability-based driving corridors can compress multi-modal traffic predictions into a small Branch MPC scenario tree while preserving safety and providing a maximum feasible decision postponing time.

desk verdict Useful practical idea for scaling Branch MPC, but the advertised safety guarantee is a heuristic backed by a good ablation and an honest limitation note, not a proven property. read the letter →

arxiv 2502.02550 v1 pith:ZJ5RJUZY submitted 2025-02-04 eess.SY cs.SY

Mohamed-Khalil Bouzidi , Bojan Derajic , Daniel Goehring , Joerg Reichardt This is my paper · ORCID
classification eess.SYcs.SY
keywords branchmodelpredictivecontrolmulti-modalmotionpredictionreachabilityanalysisdrivingcorridorscontingencyplanningdecisionpostponingautonomousscenariotree
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

This paper proposes a contingency planning framework that turns learning-based multi-modal predictions of other traffic participants into a small set of driving corridors for a Branch Model Predictive Controller. By merging corridors whose per-timestep reachable sets overlap strongly, the scenario tree stays small while every predicted mode remains represented in at least one branch. The same reachability computation provides a maximum feasible decision postponing time, so the vehicle can delay committing to a branch without making any branch infeasible. The central claim is that this reachability-based scenario selection and branching-time bound make Branch MPC scale to dense traffic while preserving safety. If correct, it would let autonomous vehicles hedge against several plausible futures at once instead of planning for only the most likely one.

What carries the argument

The central object is the driving corridor: the projection of a forward-reachable set for the ego vehicle, computed with zonotopes under a point-mass model in Frenet coordinates with discrete lane-change events, into a per-timestep position interval in the longitudinal direction and lateral bounds. Corridors are merged when their overlap measure $\Gamma$, the product over the horizon of the Jaccard index of the position sets at each timestep, reaches a threshold $\Gamma_{\min}$, with the merged corridor defined as the intersection. Feasibility of delayed branching is assessed by backpropagating the required minimum velocity to reach the longer corridor and the maximum velocity to stay in the shorter one, yielding the maximum decision postponing time $k_b^{\max}$. This machinery converts the exponential scenario selection problem into a geometric set-clustering problem with a closed-form timing bound.

What would settle it

A concrete check: construct a two-mode prediction where one mode forces the ego to accelerate hard and the other forces it to brake hard from the same initial state, while the corridor overlap product $\Gamma$ still exceeds $\Gamma_{\min}$; if the Branch MPC with the computed maximum branching time produces a trajectory that violates one corridor's constraints or finds no feasible solution, the merging criterion is insufficient.

Watch

Extended reading notes

Core claim

The paper establishes that reachability analysis can serve as the scenario-selection and timing layer for Branch MPC. For each predicted mode of each traffic participant, a driving corridor is extracted from forward-reachable sets computed with zonotopes; corridors whose overlap product, defined as the product over timesteps of the Jaccard index of their position sets, exceeds a threshold are merged by taking their intersection, and the surviving corridors become the branches of the scenario tree. The reachable sets also yield a maximum feasible branching time by backpropagating the velocity bounds between the longest and shortest corridors, and this value is compared with the adaptively estimated required branching time to decide whether a corridor must be replaced by a more conservative backup. In Monte Carlo merging scenarios, the resulting planner reported higher success rates, zero collisions, and lower jerk than single-prediction MPCC and a non-branching scenario-based baseline, while solving faster than prior Branch MPC variants.

Load-bearing premise

The load-bearing premise is that merging two driving corridors whose per-timestep reachable sets overlap, together with the computed maximum branching time, preserves a dynamically feasible contingency plan; the paper itself notes that overlap at one timestep does not imply the vehicle can satisfy both corridors' future velocity requirements.

Editorial extensions

If this is right

  • The number of branches in the scenario tree no longer grows with the number of predicted modes or traffic participants; it grows with the number of geometrically distinct driving corridors.
  • All predicted modes remain covered by at least one branch, so safety is not silently traded away by pruning low-probability predictions.
  • Collision constraint count becomes constant in the number of traffic participants ($4 \cdot N \cdot |S|$ instead of $3 \cdot |O| \cdot N \cdot |S|$), which is what makes real-time operation in dense scenes possible.
  • The maximum feasible decision postponing time provides a principled upper bound that prevents the common failure mode where a requested branching time makes all branches infeasible.
  • In the reported merging scenarios, two branches were sufficient: adding a third branch gave negligible performance gains, suggesting a minimal branch set often captures the relevant uncertainty.

Reading between the lines

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

  • An extension the paper does not pursue: the same overlap-clustering metric could be applied to scenario trees in other contingency planners, since it needs only per-timestep position sets and is predictor-agnostic.
  • A testable extension is to adapt $\Gamma_{\min}$ online, raising it when compute budget allows more branches and lowering it in dense traffic, trading conservatism against runtime.
  • The maximum postponing calculation compares only the extreme corridors by final progress; intermediate branches with conflicting velocity profiles could become infeasible after merging, a failure mode worth probing in adversarial scenarios.
  • Applying backward reachability to the intersection corridor before merging, rather than only to the final selected corridors, would catch dynamically infeasible intersections earlier.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 8 minor

Summary. The paper proposes a contingency planning framework that integrates learning-based multi-modal trajectory predictions into Branch Model Predictive Control (BMPC). The key elements are: (i) extracting driving corridors from per-mode reachable sets, (ii) pruning and clustering these corridors using a product of Jaccard overlap metrics across the horizon (Eq. 9) with pointwise intersection as the merged corridor (Eq. 10), (iii) formulating branch constraints from the resulting corridor set, and (iv) computing a maximum feasible decision-postponing time via backpropagation of longitudinal bounds between the longest and shortest corridors, with a lateral-overlap check, and replacing corridors when the required postponing time exceeds the maximum. The framework is evaluated qualitatively in an intersection scene and quantitatively in 100 Monte Carlo merging scenarios against several baselines, with reported improvements in success rate, comfort metrics, and computation time.

Significance. If the safety-preservation claim can be rigorously supported, the paper would make a useful contribution to contingency planning under multi-modal uncertainty: it attacks the computational scaling problem of Branch MPC with a principled reachability-based abstraction, and the ablation (RBMPC noMaxDP) gives concrete evidence that the maximum-postponing component is responsible for part of the observed benefit. The claimed reduction in constraint count from O(|O|*N*|S|) to O(N*|S|) is valuable, and the Monte Carlo evaluation, while not conclusive, is a reasonable first step. However, the central safety guarantee as stated in the abstract and Sec. IV-B is currently not established; the paper's own Sec. V concedes the gap between pointwise overlap and full-horizon dynamic feasibility. The framework may still be a good heuristic, and the empirical results are encouraging, but the gap between claim and proof is load-bearing.

major comments (3)
  1. [Sec. IV-B and Sec. V] The central safety-preservation claim is not established. The overlap metric in Eq. (9) and the merging rule in Eq. (10) only guarantee that the pointwise intersections D^{m1}_k ∩ D^{m2}_k are nonempty at each time step k. This does not imply the existence of a state sequence (θ_k, v_k) satisfying the dynamics (7) and actuator limits while remaining in the intersection for all k. The paper itself acknowledges in Sec. V: 'Even if there is an overlap between the driving corridor this only means that at this timestep there is one or more positions where the AV can be in both driving corridors at this time. However, this doesn't imply that it is feasible for the AV to remain within both corridors in future timesteps...' This directly contradicts the abstract's claim of 'preserving safety' and the Sec. IV-B statement that 'each driving corridor ensures constraint satisfaction for all included modes.' Please either provide a formal viability argument (for example, by applying backward reachability, as in Eq. (8), to the merged intersection set over the full horizon, rather than only to the final selected corridors) or explicitly characterize the merging step as a heuristic and rephrase the safety claims accordingly.
  2. [Sec. V and Algorithm 2] The maximum feasible branching-time calculation considers only the longest and shortest corridors by final progress (D> and D<). For |S|>2 branches, an intermediate corridor can impose a smaller feasible postponing time than either extreme, so the computed k_max_b may exceed the true maximum for all branches. The subsequent lateral-overlap check only evaluates k_max_b,longit and the following k_lc time steps, and it does not test the intermediate branches. No monotonicity or ordering argument is given to justify that checking the extremes bounds all branches. Please either extend the calculation to all branches (or provide a proof that the extreme corridors dominate the intermediate ones for the relevant feasibility condition), or conservatively take the minimum over all pairs.
  3. [Table I and Sec. VI-B] The quantitative claims of 'significantly improved safety and comfort' rest on point estimates from 100 Monte Carlo runs without confidence intervals or significance tests. Collision counts are small (0%, 1%, 2%, 6%), so the difference between, for example, RBMPC at 96% success and RBMPC noMaxDP at 94% may not be statistically meaningful. Similarly, the comfort metrics (mean velocity, jerk, steering, minimum distance) are reported as single means without standard deviations or interquartile ranges. Please report confidence intervals, standard errors, or a statistical comparison across the random seeds; this is necessary to support the 'significantly' language in the abstract and conclusion.
minor comments (8)
  1. [Sec. I] Typo: 'as as even lower probability' should be 'as even lower probability'.
  2. [Sec. II-B and Eq. (2)] The dynamic model in Eq. (2) is written as a single vector with entries 'a j \dot{\delta}', but the state vector z is defined as (x,y,ψ,v,a,δ) and the control is (j, \dot{δ}). Please clarify the exact state and control dimensions and write the differential equations for a and δ explicitly.
  3. [Sec. III] The reference 'sec. 7' in the sentence about how modes contribute to the scenario tree should be 'Sec. IV'.
  4. [Fig. 3 and Sec. IV-A] The text says 'mode 1 and mode 3 in fig. 3' but the figure shows modes 0, 1, and 2; please align the numbering.
  5. [Table I] The computation-time entries such as '88+17' are not defined in the table caption; please clarify the split (e.g., preprocessing + MPC solving time) and state the units.
  6. [Sec. IV-B] The threshold Γ_min is a free parameter with no sensitivity analysis. Since the clustering behavior and the resulting safety/conservatism trade-off depend strongly on it, please report results for at least a few plausible values.
  7. [Algorithm 1] The pseudocode's clustering loop ('while Uncompared corridors exist' and 'foreach pair') is ambiguous about whether corridors merged in the same iteration are immediately compared again; please specify the iteration order and the convergence criterion.
  8. [Conclusion] The claim that 'most predictions are effectively handled by a minimal number of driving corridors' is only illustrated for the tested scenarios; please state the empirical scope of this conclusion.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the core reachability-based derivation is self-contained and evaluated against external benchmarks.

full rationale

The paper's derivation chain is self-contained and does not reduce any of its stated contributions to its inputs by construction. Reachable sets are computed from the stated kinematic model and actuator limits (Eq. 7), and the corridor-overlap and intersection operations (Eqs. 9-10) are explicit set operations whose guarantee is not tautological: a nonempty pointwise intersection does not by itself certify a dynamically feasible whole-horizon trajectory, so the safety claim is an unsupported soundness assertion rather than a definitional equivalence. The maximum feasible branching time calculation (Sec. V) is a parameter-free backward-reachability computation from the vehicle model, independent of fitted parameters. The required decision postponing estimate and the UVD baseline come from the authors' prior work [14], but this self-citation is not load-bearing for the novel reachability-based maximum postponing computation, which is also checked by the RBMPC noMaxDP ablation; the [14] comparison is a baseline, not the derivation of the result. No parameter is fitted to the evaluation scenarios (Gamma_min is a hand-chosen design threshold), and the Monte Carlo success/collision metrics are externally measured. Hence no circular step is present; the overclaim in Sec. IV-B is a correctness/viability gap, not circularity.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The framework introduces no new physical entities. It relies on the quality and coverage of the learned predictor, on simplified ego dynamics for reachability, on the hand-tuned clustering threshold Gamma_min, and on the unproven assumption that checking only the extreme corridors bounds the maximum branching time for all branches. MPCC weights and the adaptive decision postponing estimate are inherited from prior work [23], [14].

free parameters (2)
  • Gamma_min (corridor clustering threshold)
    Hand chosen in Sec IV-B to trade off merging strength versus conservatism. No value or sensitivity analysis is reported, and it directly controls how many branches the scenario tree keeps.
  • Maximum number of scenarios considered in evaluation = 2 or 3
    When clustering outputs more corridors, Algorithm 1 keeps only the highest-probability ones. This cap is a design choice that weakens the claim that all prediction modes are preserved in the final branch set.
assumptions (4)
  • domain assumption The learned multi-modal predictor covers all relevant future behaviors of traffic participants.
    Sec II-A: the safety guarantee assumes the GMM modes from Motion Transformer include the true future. If a mode is missing, the extracted corridors do not preserve safety.
  • domain assumption A point-mass model in Frenet coordinates with at most one lane change approximates the ego vehicle's reachable sets.
    Sec IV-A: the reachable set recursion in Eq (7) uses a simplified longitudinal model and a discrete lane-change time klc. Backward refinement is skipped for efficiency.
  • domain assumption The scenario tree has a single branching point and uncertainty resolves exactly at the branching time kb.
    Sec III: the Branch MPCC formulation uses one branching point into |S| scenarios. Real uncertainty may resolve gradually, so this is a stylized model.
  • ad hoc to paper Checking only the longest and shortest corridors by final progress bounds the maximum postponing time for all branches.
    Sec V: backpropagation is performed between the first and last corridor only. No proof is given that intermediate corridors cannot impose stricter velocity or lateral constraints.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reachability-Based Contingency Planning against Multi-Modal Predictions with Branch MPC." pith.science (2026). https://pith.science/paper/ZJ5RJUZY

@misc{pith2026250202550,
  author       = {Pith},
  title        = {Pith review of: Reachability-Based Contingency Planning against Multi-Modal Predictions with Branch MPC},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZJ5RJUZY}},
  note         = {Machine review of arXiv:2502.02550}
}
read the original abstract

This paper presents a novel contingency planning framework that integrates learning-based multi-modal predictions of traffic participants into Branch Model Predictive Control (MPC). Leveraging reachability analysis, we address the computational challenges associated with Branch MPC by organizing the multitude of predictions into driving corridors. Analyzing the overlap between these corridors, their number can be reduced through pruning and clustering while ensuring safety since all prediction modes are preserved. These processed corridors directly correspond to the distinct branches of the scenario tree and provide an efficient constraint representation for the Branch MPC. We further utilize the reachability for determining maximum feasible decision postponing times, ensuring that branching decisions remain executable. Qualitative and quantitative evaluations demonstrate significantly reduced computational complexity and enhanced safety and comfort.

Figures

Figures reproduced from arXiv: 2502.02550 by the authors.

Figure 1
Figure 1. Illustration of planning in traffic scenes with uncertain [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Our Contingency Planning Framework using a Multi-modal Predictor, Reachability-based Scenario Selection, Adaptive [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. For each predicted mode of the TP, at least one [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (5 more)
Figure 4
Figure 4. Figure 4: Illustration of the collision constraint formulation for [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Snapshot with prediction of TPs and the selected [PITH_FULL_IMAGE:figures/full_fig_p006_5.png]
Figure 6
Figure 6. Figure 6: Selected driving corridors and respective maximum [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 8
Figure 8. Figure 8: Snapshot of our method during merging outputting [PITH_FULL_IMAGE:figures/full_fig_p007_8.png]
Figure 9
Figure 9. Figure 9: Distribution of comfort metric lateral and longitudinal [PITH_FULL_IMAGE:figures/full_fig_p008_9.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

24 extracted references · 21 canonical work pages

  1. [14]

    Motion planning under uncertainty: Integrating learning-based multi-modal predictors into branch model predictive control,

    M.-K. Bouzidi, B. Derajic, D. Goehring, and J. Reichardt, “Motion planning under uncertainty: Integrating learning-based multi-modal predictors into branch model predictive control,” arXiv preprint, vol. arXiv:2410.04354, 2024

    arXiv

  2. [1]

    Wayformer: Motion forecasting via simple i& efficient at- tention networks,

    N. Nayakanti, R. Al-Rfou, A. Zhou, K. Goel, K. S. Refaat, and B. Sapp, “Wayformer: Motion forecasting via simple i& efficient at- tention networks,” in 2023 IEEE International Conference on Robotics and Automation (ICRA) , 2023, pp. 2980–2987

    work page 2023

  3. [2]

    Motion transformer with global intention localization and local movement refinement,

    S. Shi, L. Jiang, D. Dai, and B. Schiele, “Motion transformer with global intention localization and local movement refinement,” Ad- vances in Neural Information Processing Systems , vol. 35, 2022

    work page 2022

  4. [3]

    Improv- ing out-of-distribution generalization of trajectory prediction for au- tonomous driving via polynomial representations,

    Y . Yao, S. Yan, D. Goehring, W. Burgard, and J. Reichardt, “Improv- ing out-of-distribution generalization of trajectory prediction for au- tonomous driving via polynomial representations,” in 2024 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) , 2024, pp. 488–495

    work page 2024

  5. [4]

    Informed spectral normalized gaussian processes for trajectory prediction,

    C. Schlauch, C. Wirth, and N. Klein, “Informed spectral normalized gaussian processes for trajectory prediction,” in ECAI 2024 - 27th European Conference on Artificial Intelligence , 2024

    work page 2024

  6. [5]

    Trajectory planning for autonomous high-speed overtaking in structured environments using robust mpc,

    S. Dixit, U. Montanaro, M. Dianati, D. Oxtoby, T. Mizutani, A. Mouzakitis, and S. Fallah, “Trajectory planning for autonomous high-speed overtaking in structured environments using robust mpc,” IEEE Transactions on Intelligent Transportation Systems , vol. 21, no. 6, pp. 2310–2323, 2020

    work page 2020

  7. [6]

    Experimental vali- dation of safe mpc for autonomous driving in uncertain environments,

    I. Batkovic, A. Gupta, M. Zanon, and P. Falcone, “Experimental vali- dation of safe mpc for autonomous driving in uncertain environments,” IEEE Transactions on Control Systems Technology , 2023

    work page 2023

  8. [7]

    Multistage Stochastic Model Predictive Control for Urban Automated Driving,

    T. Benciolini, T. Brudigam, and M. Leibold, “Multistage Stochastic Model Predictive Control for Urban Automated Driving,” in 2021 IEEE International Intelligent Transportation Systems Conference (ITSC). Indianapolis, IN, USA: IEEE, Sep. 2021, pp. 417–423

    work page 2021

Show all 24 references
  1. [8]

    Combining Stochastic and Scenario Model Predictive Control to Handle Target Vehicle Uncertainty in an Autonomous Driving Highway Scenario,

    T. Brudigam, M. Olbrich, M. Leibold, and D. Wollherr, “Combining Stochastic and Scenario Model Predictive Control to Handle Target Vehicle Uncertainty in an Autonomous Driving Highway Scenario,” in 2018 21st International Conference on Intelligent Transportation Systems (ITSC)...

    2018

  2. [9]

    Scenario-Based Decision-Making, Planning and Control for Interaction-Aware Au- tonomous Driving on Highways,

    R. Kensbock, M. Nezami, and G. Schildbach, “Scenario-Based Decision-Making, Planning and Control for Interaction-Aware Au- tonomous Driving on Highways,” in 2023 IEEE Intelligent Vehicles Symposium (IV), Jun. 2023, pp. 1–6, iSSN: 2642-7214

    2023

  3. [10]

    Interactive multi-modal motion planning with branch model predictive control,

    Y . Chen, U. Rosolia, W. Ubellacker, N. Csomay-Shanklin, and A. D. Ames, “Interactive multi-modal motion planning with branch model predictive control,” IEEE Robotics and Automation Letters , vol. 7, no. 2, pp. 5365–5372, 2022

    2022

  4. [11]

    Automated lane merging via game theory and branch model predictive control,

    L. Zhang, S. Han, and S. Grammatico, “Automated lane merging via game theory and branch model predictive control,” IEEE Transactions on Control Systems Technology , pp. 1–12, 2024

    2024

  5. [12]

    Resilient Branching MPC for Multi-Vehicle Traffic Scenarios Using Adversarial Disturbance Se- quences,

    V . Fors, B. Olofsson, and E. Frisk, “Resilient Branching MPC for Multi-Vehicle Traffic Scenarios Using Adversarial Disturbance Se- quences,” IEEE Transactions on Intelligent Vehicles , 2022

    2022

  6. [13]

    On Integrating POMDP and Scenario MPC for Planning under Uncertainty – with Applications to Highway Driving,

    C. H. Ulfsjoo and D. Axehill, “On Integrating POMDP and Scenario MPC for Planning under Uncertainty – with Applications to Highway Driving,” in 2022 IEEE Intelligent Vehicles Symposium (IV). Aachen, Germany: IEEE, Jun. 2022, pp. 1152–1160

    2022

  7. [15]

    Racp: Risk-aware contingency planning with multi-modal predictions,

    K. A. Mustafa, D. J. Ornia, J. Kober, and J. Alonso-Mora, “Racp: Risk-aware contingency planning with multi-modal predictions,” IEEE Transactions on Intelligent Vehicles, pp. 1–16, 2024

    2024

  8. [16]

    Contingency games for multi-agent inter- action,

    L. Peters, A. Bajcsy, C.-Y . Chiu, D. Fridovich-Keil, F. Laine, L. Fer- ranti, and J. Alonso-Mora, “Contingency games for multi-agent inter- action,” IEEE Robotics and Automation Letters , 2024

    2024

  9. [17]

    Lookout: Diverse multi-future prediction and planning for self-driving,

    A. Cui, S. Casas, A. Sadat, R. Liao, and R. Urtasun, “Lookout: Diverse multi-future prediction and planning for self-driving,” in Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), October 2021, pp. 16 107–16 116

    2021

  10. [18]

    Real-time capable decision making for autonomous driving using reachable sets,

    N. Kochdumper and S. Bak, “Real-time capable decision making for autonomous driving using reachable sets,” in 2024 IEEE International Conference on Robotics and Automation (ICRA) , 2024

    2024

  11. [19]

    Using reachable sets for trajectory planning of automated vehicles,

    S. Manzinger, C. Pek, and M. Althoff, “Using reachable sets for trajectory planning of automated vehicles,” IEEE Transactions on Intelligent Vehicles, vol. 6, no. 2, pp. 232–248, 2021

    2021

  12. [20]

    Computation of solution spaces for optimization-based trajectory planning,

    L. Sch ¨afer, S. Manzinger, and M. Althoff, “Computation of solution spaces for optimization-based trajectory planning,” IEEE Transactions on Intelligent Vehicles, vol. 8, no. 1, pp. 216–231, 2023

    2023

  13. [21]

    High-speed trajectory planning for autonomous vehicles using a simple dynamic model,

    F. Altch ´e, P. Polack, and A. de La Fortelle, “High-speed trajectory planning for autonomous vehicles using a simple dynamic model,” in 2017 IEEE 20th International Conference on Intelligent Transporta- tion Systems (ITSC) , 2017, pp. 1–7

    2017

  14. [22]

    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 et al. , “Large scale interactive motion forecasting for autonomous driving: The waymo open motion dataset,” in Proceedings of the IEEE/CVF International Conference on Computer ...

    2021

  15. [23]

    Model Predictive Contouring Control for Collision Avoidance in Unstructured Dynamic Environments,

    B. Brito, B. Floor, L. Ferranti, and J. Alonso-Mora, “Model Predictive Contouring Control for Collision Avoidance in Unstructured Dynamic Environments,” IEEE Robotics and Automation Letters , vol. 4, no. 4, pp. 4459–4466, 2019

    2019

  16. [24]

    Self-driving like a human driver instead of a robocar: Personalized comfortable driving experience for autonomous vehicles,

    I. Bae, J. Moon, J. Jhung, H. Suk, T. Kim, H. Park, J. Cha, J. Kim, D. Kim, and S. Kim, “Self-driving like a human driver instead of a robocar: Personalized comfortable driving experience for autonomous vehicles,” 2022

    2022

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.