Pith. sign in

REVIEW 4 major objections 3 minor 27 references

Dual Control for Interactive Autonomous Merging with Model Predictive Diffusion

T0 review · 4 major / 3 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read This paper claims that planning over predicted beliefs about driver intent lets an autonomous vehicle actively probe traffic and complete highway merges faster than passive planners.

desk verdict A real hardware contribution with a broken proof of the central predicted-belief theorem; the paper should be reviewed as an engineering report after the theory is corrected or explicitly reframed as an approximation. read the letter →

arxiv 2502.09918 v1 pith:LSJ5VTWJ submitted 2025-02-14 cs.RO cs.SYeess.SYmath.OC

Jacob Knaup , Jovin D'sa , Behdad Chalaki , Hossein Nourkhiz Mahjoub , Ehsan Moradi-Pari , Panagiotis Tsiotras This is my paper · ORCID
classification cs.ROcs.SYeess.SYmath.OC
keywords dualcontrolmodelpredictivediffusionautonomousmergingbeliefpropagationactivelearningBayesianinferencehuman-robotinteractionF1-Tenth
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper proposes a dual-control framework in which an autonomous vehicle plans a highway merge while actively reducing uncertainty about other drivers' intentions, rather than treating prediction and planning as separate steps. The authors derive a recursive update for predicted belief distributions over driver behavior parameters and fold that update into the receding-horizon optimization, so the chosen control trajectory is partly aimed at probing which driver will yield. To solve the resulting nonconvex, collision-penalized problem online, they introduce a model-based diffusion sampler that reuses the previous time step's solutions as a multimodal prior. In F1-Tenth hardware experiments with a human-controlled yielding vehicle, the method completed every merge within 15 meters, with an average merge distance of 4.3 meters, versus 7.1 meters for a dual-control MPPI baseline and a 58% success rate for a passive-learning baseline.

What carries the argument

The load-bearing objects are (i) the predicted-belief recursion of Theorem 1, Eq. (9), which propagates the posterior over driver parameters conditioned on the planned control sequence rather than on unobserved future observations, and (ii) the model predictive diffusion sampler, which forms a multimodal prior by corrupting the previous time step's optimal control samples (modes) and reverses a truncated diffusion SDE using an explicit score computed from the cost distribution. The recursion is what imports active learning into the optimization; the diffusion sampler is what makes the nonconvex, collision-penalized cost tractable online. A particle-filter approximation in Eqs. (34)-(35c) turns both the belief update and the predicted-belief propagation into weighted sample averages.

What would settle it

Run a two-step simulation with a nonlinear dynamics model where $x_{k+1}$ depends on $x_k$ and the same latent $\theta$ appears in every step; compute the left and right sides of Corollary 1 by Monte Carlo over full trajectories and compare the resulting predicted beliefs at each horizon time. A measurable gap in the predicted weights or in downstream merge decisions would show the recursion is approximate.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central claim is that the future belief about an unobserved driver-behavior parameter $\theta$ can be predicted online by the recursive update $\hat{b}_{k+1|t}(\theta) \propto \hat{b}_{k|t}(\theta) \, \mathbb{E}_{\bar{\theta} \sim b_t}[f(x_{k+1}|x_t,u_{t:k},\theta) \,|\, x_t,u_{t:k},\bar{\theta}]$, and that replacing a fixed-belief stochastic MPC problem with this predicted-belief dynamics produces a genuine dual-control policy: the planned control sequence shapes the entropy of the future belief distribution, so the vehicle actively probes other drivers. This is paired with a model-based diffusion solver that maintains multiple modes from the previous time step's solutions as a dynamic prior and denoises them with an explicit score function, turning the nonconvex, penalty-laden merge cost into a sampling problem solvable at 10 Hz on embedded hardware.

Load-bearing premise

The proof of the predicted-belief update assumes that the expectation over future trajectories can be pulled inside the product of likelihood terms, an exchange that does not hold for general nonlinear dynamics and that makes Eq. (9) an approximation rather than an exact recursive posterior.

Editorial extensions

If this is right

  • A passive predict-then-act controller (EMPPI) completed the merge in only 58% of 12 hardware trials, whereas both dual-control methods completed all trials, indicating that planning over belief dynamics is what enables reliable merging.
  • The proposed diffusion solver completed merges at an average distance of 4.3 meters versus 7.1 meters for dual MPPI, implying that the multimodal dynamic prior yields solutions closer to the global optimum under the same real-time budget.
  • Because the method ran at 10 Hz on an embedded Jetson Orin Nano, the approach is compatible with real-time onboard planning rather than offline trajectory optimization.
  • The human-yielding experiments show the framework tolerates model mismatch: the traffic car's behavior was not generated by the MR-IDM model assumed in the belief prior, yet the ego vehicle still inferred and exploited the friendly driver.

Reading between the lines

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

  • Editorial inference: Because Theorem 1's derivation exchanges an expectation over future trajectories with a product of per-step likelihood factors, a natural testable extension is to measure how much Eq. (9) diverges from full-trajectory Monte Carlo belief propagation as the horizon and the coupling between vehicles grow.
  • Editorial inference: The multimodal dynamic prior should transfer to other online nonconvex trajectory optimization problems, such as navigation in dense clutter or off-road driving, where warm-starting from previous solution modes can shrink the number of denoising steps.
  • Editorial inference: The human-yielding hardware trials suggest the framework could estimate higher-level social preferences directly, rather than identifying parameters of a fixed car-following model, which would extend it to more diverse human behaviors.
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

4 major / 3 minor

Summary. The paper proposes a dual control framework for autonomous highway merging. The ego vehicle maintains a particle-filter belief over parameters of other drivers' behavior and solves a receding-horizon stochastic optimal control problem whose cost includes a predicted belief that is meant to account for future information gain. The optimization is performed by a new model-based diffusion solver that uses previous solutions as a multimodal prior. The method is evaluated in F1-Tenth hardware experiments with a human-controlled yielding vehicle, reporting higher merge success and shorter merge distance than two MPPI baselines. The abstract and contributions emphasize that the predicted belief distributions are 'rigorously derived,' and this derivation is the theoretical centerpiece of the paper.

Significance. If the theoretical derivation were correct, the paper would make a useful contribution: it extends dual control beyond linear/Gaussian assumptions, proposes a diffusion-based MPC solver with a receding-horizon prior, and provides real-time hardware validation. The experiments are a genuine strength: 10 Hz operation on an embedded platform, 12 trials per method, and a human-driven yielding component that introduces model mismatch. However, the central claim of a rigorously derived predicted belief distribution is not supported: Theorem 1 is internally inconsistent and its proof is invalid. Because this theorem is the basis for the dual-control formulation and the claimed active-learning benefit, the manuscript's main theoretical contribution fails. The empirical comparison alone may support a weaker claim that a heuristic predicted-belief update combined with diffusion-based sampling works well in this testbed, but that is not what the paper currently argues.

major comments (4)
  1. [Section III-A, Eqs. (8)-(9)] The object defined in (8) is not the object computed in (9). Since u_{t+1:k} are planned controls generated by the ego policy from information available at time t, they carry no additional information about θ given ξ_{0:t}, so b(θ|ξ_{0:t}, u_{t+1:k}) = b_t(θ). Equation (9), by contrast, averages the posterior that would arise from hypothetical future observations over their predictive distribution; that average is a mixture of posteriors, not the conditional distribution in (8). The proof implicitly replaces the conditioning in (8) with this mixture, so the theorem does not establish (9).
  2. [Proof of Theorem 1 and Corollary 1] The derivation factors the product of Markov transition densities. The proof writes the joint integral over x_{t+1:k+1} as a product of integrals over each x_{ℓ+1}, and Corollary 1 moves the expectation inside the product. For the nonlinear Markov system (1), the trajectory likelihood ∏_{ℓ=t}^k f(x_{ℓ+1}|x_ℓ,u_ℓ,θ) does not factor in this way, because each x_{ℓ+1} is correlated with x_ℓ and with θ through the preceding states; an expectation of the product is not the product of the expectations in general. Consequently, Eq. (9) is at best an unstated approximation, not a rigorous recursive posterior update.
  3. [Proof of Theorem 1] The displayed integral is not a well-formed mathematical expression. It contains b_t(θ) both outside and inside the integration over θ, the integration measures are ambiguous (the expression 'dθdxt+1' does not match the nested integral signs), and the domain of the θ integral is unspecified. This makes the proof impossible to verify as written and further indicates that (9) is obtained by an invalid interchange of integration and product.
  4. [Section III-C, Eq. (34)] The particle update (34) is not a particle approximation of (9). In (9) the expectation is over θ̄∼b_t of the transition density evaluated at the fixed parameter θ, whereas in (34) the density is evaluated at the expected state and at the sampled parameter θ̂_j, with the expectation moved inside the pdf. The text labels this as a computational shortcut, but no error bound or consistency argument is given. Since Theorem 4's dual-control conclusion and the cost approximation (35) rely on (34), the algorithmic claim of active uncertainty reduction is not supported by the stated theory.
minor comments (3)
  1. [Section IV, Eq. (38)] The notation Q(⋆) is undefined; it should be written as (x^0_k − x_g)ᵀ Q (x^0_k − x_g).
  2. [Section IV, Eq. (37)] The constraint 'δmin ≤ δt ≤ amax' should read 'δmin ≤ δt ≤ δmax'.
  3. [Algorithm 1] The loop starts at t = 1, but the initialization block refers to {u^{0,j}_{-1}}, which is never used inside the loop; the indexing of the bootstrap samples should be clarified.

Circularity Check

2 steps flagged · score 6.0 of 10

The predicted-belief derivation in Theorem 1 factors a product of correlated transition likelihoods as a product of expectations, and Theorem 4's dual-control effect is asserted from the very weight update that defines it; the central active-learning claim is therefore partly circular/by construction rather than rigorously derived.

  1. other [Section III-A, Theorem 1 and Corollary 1, Eqs. (8)-(10)]
    "However, since the future states are not observed, we expand (8) using the predictive distribution marginalized over the unknown future observations given by ˆbk+1|t(θ) ∝ bt(θ) R R Qk ℓ=t f(xℓ+1|xℓ,uℓ,θ)bt(θ)dθdxt+1 ... which is concisely expressed in a recursive form by (9)."

    Eq. (8) defines the predicted belief as p(θ|ξ0:t,u_{t+1:k}); since u_{t+1:k} are planned controls known at time t, that conditioning carries no new information about θ. Theorem 1 instead computes an expectation over unobserved future states. The displayed proof then replaces the integral of the product ∏f over the Markov trajectory by the product of integrals/expectations. That equality is exactly the conditional-independence claim needed to make (9) true, and it is not supplied by dynamics (1). Thus Eq. (9) is not derived from Eq. (8); the recursive belief update is an unstated ansatz presented as a rigorously derived result.

  2. self definitional [Section III-C, Theorem 4, after Eq. (35)]
    "Theorem 4 ([22]). The solution to problem (35) preserves the dual control effect [9], that is, the planned control actions affect the entropy of the predicted future belief distribution. Proof. The proof follows from (34), in which the planned control sequence ut:k|t affects the weights ˆωj k+1|t of the categorical distribution over {θj}."

    The dual-control/probing property is not derived from an external principle; it is built into (34)/(35c), where the belief weights are updated using the planned controls. The theorem states that planned controls enter the weight update, which is true by definition of the update. Therefore the central qualitative claim that the planner actively reduces uncertainty is guaranteed by construction rather than predicted from first principles.

full rationale

The paper contains genuine external content: hardware F1-Tenth experiments, comparisons with DMPPI and EMPPI, and a model-based diffusion solver extending external work. Self-citations [22] and [23] are present but not uniquely load-bearing. However, the theoretical centerpiece advertised in the abstract—'rigorously derive predicted belief distributions'—is undercut by the factorization in Theorem 1/Corollary 1, which assumes the conditional independence that Eq. (9) needs, and the 'dual control effect' theorem is a tautological restatement of the control-dependent weight update. These two issues make the active-learning claim partially circular/by construction, so a score of 6 is appropriate. If Theorem 1 were explicitly rewritten as an approximation rather than a rigorous derivation, the circularity score would drop substantially.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The framework rests on the assumed traffic dynamics (MR-IDM), full observability, and a number of hand-chosen cost and diffusion parameters. The most fragile entry is the ad hoc factorization of expected future likelihoods in the predicted belief update, which the paper treats as a theorem. The bootstrap prior relies on consecutive optima being close, a standard MPC assumption. No code or data are provided, so the listed free parameters are not fully specified.

free parameters (4)
  • Cost weights Q, Qf, R, Qpen = not reported
    Shape the trade-off between goal progress, control effort, collision penalty, and road violations; chosen by hand and not numerically listed.
  • Diffusion schedule {A_tau, B_tau} and number of steps N_d = not reported
    Forward noise schedule and denoising depth affect solution quality; chosen to hit 10 Hz real-time and not specified.
  • Sample counts N_p, N_hat_p, N_m, N_s = not reported
    Particle, mode, and importance-sample counts are tuned for real-time performance and not stated.
  • Process noise covariance Sigma_w = not reported
    Added to the dynamics in Eq. (37); the value is not reported.
assumptions (5)
  • domain assumption State x_t is fully observable for all vehicles.
    Stated in Section IV; the belief is over behavior parameters only, not over states.
  • domain assumption Traffic vehicle accelerations follow the MR-IDM model with fixed unknown parameters theta.
    Used in Eq. (37); the hardware experiments intentionally violate this for the friendly vehicle, making the model misspecified by design.
  • ad hoc to paper The expected product of future likelihoods equals the product of expected likelihoods in the predicted belief update.
    Required for Theorem 1 and Corollary 1 in Section III-A; not justified and generally false for nonlinear dynamics.
  • domain assumption Optimal control sequences are similar between consecutive time steps.
    Used in Eq. (29) to construct the multimodal dynamic prior; a standard MPC warm-start assumption.
  • domain assumption A particle filter without resampling is sufficient for belief propagation.
    Remark 3 in Section III-C; carried over from [22], not established for these experiments.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Dual Control for Interactive Autonomous Merging with Model Predictive Diffusion." pith.science (2026). https://pith.science/paper/LSJ5VTWJ

@misc{pith2026250209918,
  author       = {Pith},
  title        = {Pith review of: Dual Control for Interactive Autonomous Merging with Model Predictive Diffusion},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LSJ5VTWJ}},
  note         = {Machine review of arXiv:2502.09918}
}
read the original abstract

Interactive decision-making is essential in applications such as autonomous driving, where the agent must infer the behavior of nearby human drivers while planning in real-time. Traditional predict-then-act frameworks are often insufficient or inefficient because accurate inference of human behavior requires a continuous interaction rather than isolated prediction. To address this, we propose an active learning framework in which we rigorously derive predicted belief distributions. Additionally, we introduce a novel model-based diffusion solver tailored for online receding horizon control problems, demonstrated through a complex, non-convex highway merging scenario. Our approach extends previous high-fidelity dual control simulations to hardware experiments, which may be viewed at https://youtu.be/Q_JdZuopGL4, and verifies behavior inference in human-driven traffic scenarios, moving beyond idealized models. The results show improvements in adaptive planning under uncertainty, advancing the field of interactive decision-making for real-world applications.

Figures

Figures reproduced from arXiv: 2502.09918 by the authors.

Figure 1
Figure 1. Snapshots of traffic merge experiment at 8-second increments, when the ego vehicle must overtake (left) or yield [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Belief prediction approach enabling active learning. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Model Predictive Diffusion: In the forward pass, Nm samples from the previous time-step’s optimal distribution s 0 (·|xt) are corrupted with noise to form the dynamic prior distribution for the current time-step. In the backward pass, samples are drawn from the prior distribution s Nd (·|xt+1) and are denoised to construct samples from the updated optimal distribution s 0 (·|xt+1). The samples u τ,j t:t+N−1 are inde… view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: Merge scenario [PITH_FULL_IMAGE:figures/full_fig_p009_4.png]
Figure 5
Figure 5. Figure 5: F1-Tenth platform. The proposed approach is compared with two baseline methods: the dual model predictive path integral control (DMPPI) from [22] and the ensemble MPPI (EMPPI) from [27]. DMPPI solves the proposed problem (35) while ablat￾ing the multi-modal component o…
Figure 6
Figure 6. Figure 6: Depiction of single trial for proposed and baseline methods. [PITH_FULL_IMAGE:figures/full_fig_p012_6.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 18 canonical work pages

  1. [1]

    Model predic- tive control for autonomous ground vehicles: a review,

    S. Yu, M. Hirche, Y . Huang, H. Chen, and F. Allg¨ower, “Model predic- tive control for autonomous ground vehicles: a review,” Autonomous Intelligent Systems, vol. 1, pp. 1–17, 2021

    work page 2021

  2. [2]

    Survey of deep reinforcement learning for motion planning of autonomous vehicles,

    S. Aradi, “Survey of deep reinforcement learning for motion planning of autonomous vehicles,”IEEE Transactions on Intelligent Transporta- tion Systems, vol. 23, no. 2, pp. 740–759, 2020

    work page 2020

  3. [3]

    Multi-robot cooperative navigation in crowds: A game-theoretic learning-based model predictive control approach,

    V .-A. Le, V . Tadiparthi, B. Chalaki, H. N. Mahjoub, J. D’sa, E. Moradi- Pari, and A. A. Malikopoulos, “Multi-robot cooperative navigation in crowds: A game-theoretic learning-based model predictive control approach,” in IEEE International Conference on Robotics and Automa- tion (ICRA), Yokohama, Japan, May 13–17, 2024, pp. 4834–4840. Fig. 6: Depiction o...

    work page 2024

  4. [4]

    A hybrid deep re- inforcement learning and optimal control architecture for autonomous highway driving,

    N. Albarella, D. G. Lui, A. Petrillo, and S. Santini, “A hybrid deep re- inforcement learning and optimal control architecture for autonomous highway driving,” Energies, vol. 16, no. 8, p. 3490, 2023

    work page 2023

  5. [5]

    Decision- making based on reinforcement learning and model predictive control considering space generation for highway on-ramp merging,

    H. Kimura, M. Takahashi, K. Nishiwaki, and M. Iezawa, “Decision- making based on reinforcement learning and model predictive control considering space generation for highway on-ramp merging,” IFAC- PapersOnLine, vol. 55, no. 27, pp. 241–246, 2022

    work page 2022

  6. [6]

    Interaction-aware trajectory planning for autonomous vehicles with analytic integration of neural networks into model predictive control,

    P. Gupta, D. Isele, D. Lee, and S. Bae, “Interaction-aware trajectory planning for autonomous vehicles with analytic integration of neural networks into model predictive control,” in IEEE International Con- ference on Robotics and Automation (ICRA) , London, UK, May 29– June 2, 2023, pp. 7794–7800

    work page 2023

  7. [7]

    Active probing and influencing human behaviors via autonomous agents,

    S. Wang, Y . Lyu, and J. M. Dolan, “Active probing and influencing human behaviors via autonomous agents,” in IEEE International Con- ference on Robotics and Automation (ICRA) , London, UK, May 29– June 2, 2023, pp. 1514–1521

    work page 2023

  8. [8]

    Stochastic MPC with Dual Control for Autonomous Driving with Multi-Modal Interaction-Aware Predictions

    S. H. Nair, V . Govindarajan, T. Lin, Y . Wang, E. H. Tseng, and F. Borrelli, “Stochastic MPC with dual control for autonomous driv- ing with multi-modal interaction-aware predictions,” arXiv preprint arXiv:2208.03525, 2022

    work page Pith review arXiv 2022

Show all 27 references
  1. [9]

    Active uncertainty reduction for human- robot interaction: An implicit dual control approach,

    H. Hu and J. F. Fisac, “Active uncertainty reduction for human- robot interaction: An implicit dual control approach,” in International Workshop on the Algorithmic Foundations of Robotics. College Park, MD: Springer, June 22–24, 2022, pp. 385–401

    2022

  2. [10]

    Active uncertainty reduction for safe and efficient interaction planning: A shielding-aware dual control approach,

    H. Hu, D. Isele, S. Bae, and J. F. Fisac, “Active uncertainty reduction for safe and efficient interaction planning: A shielding-aware dual control approach,” The International Journal of Robotics Research , vol. 43, no. 9, pp. 1382–1408, 2024

    2024

  3. [11]

    Plan- ning with diffusion for flexible behavior synthesis,

    M. Janner, Y . Du, J. B. Tenenbaum, and S. Levine, “Plan- ning with diffusion for flexible behavior synthesis,” arXiv preprint arXiv:2205.09991, 2022

    2022 arXiv

  4. [12]

    Diffusion policy: Visuomotor policy learning via ac- tion diffusion,

    C. Chi, Z. Xu, S. Feng, E. Cousineau, Y . Du, B. Burchfiel, R. Tedrake, and S. Song, “Diffusion policy: Visuomotor policy learning via ac- tion diffusion,” The International Journal of Robotics Research , p. 02783649241273668, 2023

    2023

  5. [13]

    Conformal prediction for uncertainty-aware planning with dif- fusion dynamics model,

    J. Sun, Y . Jiang, J. Qiu, P. Nobel, M. J. Kochenderfer, and M. Schwa- ger, “Conformal prediction for uncertainty-aware planning with dif- fusion dynamics model,” Advances in Neural Information Processing Systems, vol. 36, Dec. 10–15, 2024

    2024

  6. [14]

    Learning a diffusion model policy from rewards via q-score matching,

    M. Psenka, A. Escontrela, P. Abbeel, and Y . Ma, “Learning a diffusion model policy from rewards via q-score matching,” arXiv preprint arXiv:2312.11752, 2023

    2023 arXiv

  7. [15]

    Denoising diffusion probabilistic models,

    J. Ho, A. Jain, and P. Abbeel, “Denoising diffusion probabilistic models,” Advances in Neural Information Processing Systems, vol. 33, pp. 6840–6851, Dec. 6–12, 2020

    2020

  8. [16]

    Generative modeling by estimating gradients of the data distribution,

    Y . Song and S. Ermon, “Generative modeling by estimating gradients of the data distribution,” Advances in Neural Information Processing Systems, vol. 32, Dec. 8–14, 2019

    2019

  9. [17]

    Model-based diffusion for trajectory optimization,

    C. Pan, Z. Yi, G. Shi, and G. Qu, “Model-based diffusion for trajectory optimization,” arXiv preprint arXiv:2407.01573 , 2024

    2024 arXiv

  10. [18]

    Diffusion models: A comprehensive survey of methods and applications,

    L. Yang, Z. Zhang, Y . Song, S. Hong, R. Xu, Y . Zhao, W. Zhang, B. Cui, and M.-H. Yang, “Diffusion models: A comprehensive survey of methods and applications,” ACM Computing Surveys, vol. 56, no. 4, pp. 1–39, 2023

    2023

  11. [19]

    Score-based generative modeling through stochastic differential equations,

    Y . Song, J. Sohl-Dickstein, D. P. Kingma, A. Kumar, S. Ermon, and B. Poole, “Score-based generative modeling through stochastic differential equations,” arXiv preprint arXiv:2011.13456 , 2020

    2011 arXiv

  12. [20]

    A survey on the coordination of connected and automated vehicles at intersections and merging at highway on-ramps,

    J. Rios-Torres and A. A. Malikopoulos, “A survey on the coordination of connected and automated vehicles at intersections and merging at highway on-ramps,” IEEE Transactions on Intelligent Transportation Systems, vol. 18, no. 5, pp. 1066–1077, 2016

    2016

  13. [21]

    Highway on-ramp merging for mixed traffic: Recent advances and future trends,

    S. A. Fernandez, M. A. Marinho, M. Vakilzadeh, and A. Vinel, “Highway on-ramp merging for mixed traffic: Recent advances and future trends,” in IEEE 29th International Conference on Network Protocols (ICNP), Dallas, TX, Nov. 1–5, 2021, pp. 1–6

    2021

  14. [22]

    Active learning with dual model predictive path-integral control for interaction-aware autonomous highway on- ramp merging,

    J. Knaup, J. D’sa, B. Chalaki, T. Naes, H. N. Mahjoub, E. Moradi- Pari, and P. Tsiotras, “Active learning with dual model predictive path-integral control for interaction-aware autonomous highway on- ramp merging,” in IEEE International Conference on Robotics and Automation (I...

    2024

  15. [23]

    MR-IDM–merge reactive intelligent driver model: Towards enhancing laterally aware car-following models,

    D. Holley, J. D’sa, H. N. Mahjoub, G. Ali, B. Chalaki, and E. Moradi- Pari, “MR-IDM–merge reactive intelligent driver model: Towards enhancing laterally aware car-following models,” arXiv preprint arXiv:2305.12014, 2023

    2023 arXiv

  16. [24]

    JAX: composable transformations of Python+NumPy programs,

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman- Milne, and Q. Zhang, “JAX: composable transformations of Python+NumPy programs,” 2018. [Online]. Available: http://github. com/google/jax

    2018

  17. [25]

    Kinematic and dynamic vehicle models for autonomous driving control design,

    J. Kong, M. Pfeiffer, G. Schildbach, and F. Borrelli, “Kinematic and dynamic vehicle models for autonomous driving control design,” in IEEE Intelligent Vehicles Symposium (IV) , Seoul, South Korea, June 28–July 1, 2015, pp. 1094–1099

    2015

  18. [26]

    Information-theoretic model predictive control: Theory and applica- tions to autonomous driving,

    G. Williams, P. Drews, B. Goldfain, J. M. Rehg, and E. A. Theodorou, “Information-theoretic model predictive control: Theory and applica- tions to autonomous driving,” IEEE Transactions on Robotics, vol. 34, no. 6, pp. 1603–1622, 2018

    2018

  19. [27]

    Model-based generalization under parameter uncertainty using path integral control,

    I. Abraham, A. Handa, N. Ratliff, K. Lowrey, T. D. Murphey, and D. Fox, “Model-based generalization under parameter uncertainty using path integral control,” IEEE Robotics and Automation Letters , vol. 5, no. 2, pp. 2864–2871, 2020

    2020

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.