Pith. sign in

REVIEW 3 major objections 6 minor 33 references

Frenet Corridor Planner: An Optimal Local Path Planning Framework for Autonomous Driving

T0 review · 3 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

Pith's one-line read The Frenet Corridor Planner claims that local path planning for autonomous driving reduces to a corridor-constrained optimization in Frenet space, solving in tens of milliseconds and producing smoother, safer paths than A*, RRT*, and…

desk verdict Useful, efficient local planner with a real gap in the Frenet kinematics on curved roads; deserves revision, not desk rejection. read the letter →

arxiv 2505.03695 v1 pith:DCXV24LI submitted 2025-05-06 cs.RO cs.SYeess.SY

classification cs.ROcs.SYeess.SY
keywords autonomousdrivinglocalpathplanningFrenetframecorridor-basedkinematicbicyclemodeltrajectoryoptimizationpath-speeddecompositionobstacleavoidance
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 introduces the Frenet Corridor Planner (FCP), a local path planning method for autonomous driving that formulates obstacle avoidance as an optimization problem inside a lateral corridor in the Frenet frame. Its central claim is that this decomposition makes real-time smooth path planning tractable: instead of solving a non-convex collision-avoidance problem directly, FCP first builds conservative corridor bounds from obstacle geometry, then optimizes a path that stays inside them while minimizing smoothness, steering effort, boundary proximity, and dynamic-obstacle risk. The paper reports an average solve time around 0.035–0.042 s, and in a scenario with parked cars and an oncoming vehicle FCP is the only planner among A*, RRT*, and B-RRT* that completes the maneuver, with the smallest maximum yaw change. Validation in high-fidelity simulation and on scaled autonomous vehicles supports the claim that the approach transfers from idealized planning to noisy, real-time hardware settings. If correct, FCP offers a modular, low-latency path layer that can sit between a global route and a speed planner in an autonomous driving stack.

What carries the argument

The load-bearing mechanism is the lateral corridor in Frenet space together with a modified space-domain bicycle kinematics model. The corridor is produced by a single-pass boundary generation algorithm that scans obstacle points at the queried longitudinal positions and sets the lower and upper lateral bounds, so the optimizer only ever sees bound constraints rather than non-convex collision constraints. The kinematics model removes time and speed, fixes a constant longitudinal step Δs, and uses the projection Δs = Δl cos(ψ + β) to write the bicycle update directly in the Frenet variables: d_{k+1} = d_k + tan(ψ_k + u_k)Δs and ψ_{k+1} = ψ_k + (Δs/ℓ_r) sin(u_k)/cos(ψ_k + u_k), with u_k a linearized proxy for the slip angle. The optimizer also restricts the steering set using the reference path's curvature, includes a bounded slack variable for feasibility under perception noise, and adds an inverse-square penalty around predicted dynamic-obstacle positions. This combination is what lets the planner produce a smooth, kinematically feasible path inside the corridor at low computational cost.

What would settle it

On a constant-curvature reference path with lateral offset d comparable to the path's radius, compare the vehicle's actual Frenet s-advance against the paper's Δs = Δl cos(ψ + β); if the planned path systematically violates the corridor bounds by more than the slack variable permits as κd grows, the projection assumption is falsified.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that optimal local path planning can be made both smooth and fast by separating the problem into two stages: first, a conservative geometric corridor in Frenet coordinates that encodes where the vehicle may drive; second, a small optimization whose only non-convexity is the kinematics model. The corridor is built by modeling vehicles as safety-augmented bounding boxes and pedestrian clusters as convex hulls, then deciding, for each static obstacle, whether it constrains the lower or upper boundary of the drivable region. The optimizer then minimizes a convex cost over lateral deviation and steering input, subject to the corridor as simple bound constraints, while a slack variable preserves feasibility under perception noise and dynamic obstacles enter as additive risk penalties. Because collision avoidance is encoded in the bounds rather than in inequality constraints, the optimization remains small enough to solve in roughly 0.035 to 0.042 seconds, and the reported scenario results show FCP completing a maneuver that none of the sampling or graph baselines completed.

Load-bearing premise

Everything rests on treating Δs = Δl cos(ψ + β) as an adequate Frenet projection of the bicycle model onto the reference path, without the usual curvature-dependent (1 − κd) correction.

Editorial extensions

If this is right

  • FCP's reported runtime figures make it usable for closed-loop local replanning at typical automotive control rates, not just offline path generation.
  • Encoding obstacles as corridor bounds removes the need for non-convex collision constraints, so adding more static obstacles changes only the bound profiles, not the optimization structure.
  • Lower maximum yaw change and lower angular jerk than the baselines imply smoother steering commands downstream, which can reduce passenger discomfort and actuator wear.
  • Because the path layer is modular and the speed planner is separate, adopting FCP does not require replacing an existing speed planner or downstream controller.

Reading between the lines

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

  • Editorial extension: the corridor construction depends only on a reference curve and obstacle edges, so the same machinery could be lifted to parking lots or off-road paths by supplying an arbitrary drivable reference and re-running the single-pass boundary algorithm.
  • Editorial extension: the reported runtime comes from a Python/CasADi implementation on one core, so a compiled rewrite would likely push the planner below 20 ms, enabling higher replanning rates or longer horizons.
  • Editorial extension: the missing (1 − κd) transport term in the Frenet projection is the natural stress test; a simple experiment on a constant-curvature loop with known obstacle offsets would reveal whether the steering-bound correction is sufficient for tight bends.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 6 minor

Summary. The paper presents the Frenet Corridor Planner (FCP), a modular local path-planning framework for autonomous driving. FCP converts perception data into safety-augmented bounding boxes and convex hulls in a Frenet frame, uses a decision governor and boundary generator to build a drivable corridor, and then solves a nonlinear optimization problem with a space-domain bicycle kinematic model to produce a path that balances smoothness, boundary clearance, and dynamic-obstacle risk. The planned path is passed to an external speed planner. The authors validate FCP against A*, RRT*, and B-RRT* in a scenario-based comparison, in CARLA Monte Carlo trials, in a runtime study, and on 1/10-scale hardware, reporting average runtimes around 0.035-0.042 s and favorable smoothness metrics.

Significance. If the kinematic model is correct, FCP is a practically valuable contribution: it gives a real-time, optimization-based local path planner with a clean corridor formulation, and the validation effort is substantial, including simulation, Monte Carlo trials, and a hardware demonstration. The paper also contains a self-contained proof of a steering-angle approximation bound (Section III-D.1), and the overall pipeline is modular enough to be reused with other speed planners and controllers. However, the central correctness claim depends on the Frenet-space kinematics, and the manuscript's treatment of curvature is incomplete: the exact Frenet transport terms involving the (1-κd) factor are dropped, and the proposed curvature correction only modifies steering bounds. Because of this, the claims of kinematic feasibility, corridor conformance, and safety on curved roads are not established as written.

major comments (3)
  1. [III-D.1, Eqs. (25)-(27)] The kinematics model in Eqs. (25)-(27) omits the (1-κd) factor of exact Frenet transport. For a curved reference path q(s)=R(s)+d(s)N(s), one has q'(s)=(1-κd)T+d'N, so the exact relations are ds=Δl cos(ψ+β)/(1-κd), d_{k+1}=d_k+(1-κd)tan(ψ+β)Δs, and ψ_{k+1}=ψ_k+(1-κd)(Δs/ℓ_r)sin(β)/cos(ψ+β). The paper instead uses Δs=Δl cos(ψ+β), d_{k+1}=d_k+tan(ψ+β)Δs, and ψ_{k+1}=ψ_k+(Δs/ℓ_r)sin(β)/cos(ψ+β). Remark 4 acknowledges the incompatibility, but the correction in Section III-D.2 only tightens steering bounds using the reference heading change; it does not correct the dynamics. The omitted factor is not negligible: with κ=0.05 1/m and d=2 m, it equals 0.9, a 10% kinematic error. Since the hardware scenario explicitly includes a curved section, the claims of corridor conformance, kinematic feasibility, and safety on curved roads are not supported by the current model. I recommend either deriving and using the exact Frenet kinematics, or explicitly restricting the claims to low-curvature references and adding experiments that isolate the curved-road regime.
  2. [III-D.2, Eq. (21)] The claim that Δψ_k is lower bounded by (Δs/ℓ_r)tan(u_k) for |ψ_k+u_k|<π/2 is not true in general. For example, with ψ_k=-0.5 rad and u_k=0.3 rad, cos(ψ_k+u_k)=cos(-0.2)≈0.980 > cos(0.3)≈0.955, so sin(u_k)/cos(ψ_k+u_k)≈0.301 < tan(0.3)≈0.309; the inequality is reversed. This matters because the paper uses the bound to argue that the steering restriction in (23) over-constrains the feasible set. If the bound fails, the curvature-induced actuation limit does not necessarily ensure kinematic feasibility with respect to the reference curvature. Please either provide a correct derivation of a valid bound or remove the claim that the feasible set is over-constrained.
  3. [III-D.4, Abstract] The paper's abstract and introduction state that FCP 'ensures smooth and safe navigation around obstacles,' but the path optimizer handles dynamic obstacles only as an additive penalty in Eq. (31); hard safety for dynamic obstacles is deferred to the speed planner, which 'guarantees safety in space-time' per Section III-D.4. As written, the safety claim for FCP itself is therefore overbroad and the dynamic-obstacle safety guarantee is not proven in this manuscript. Please either soften the safety claims to refer to the full pipeline, or provide a concrete end-to-end safety analysis that includes the speed planner and control layer.
minor comments (6)
  1. [III-D.1, Eq. (20)] The inequality in Eq. (20) appears to be missing a factor: the proof of Eq. (14) gives arctan(a tan δ) ≥ aδ, not arctan(a tan δ) ≥ δ. Please correct the displayed inequality and the surrounding sentence.
  2. [III-B] The partition notation is inconsistent: the text writes Ot = Oub_t ∪ O ub_t and Oub_t ∩ O ub_t = ∅. The second set should presumably be the lower-bound set Olb_t, not a repeat of the upper-bound set.
  3. [IV-B, Table I] The row labels in Table I are not fully consistent with the metric definitions in Section IV-B: the text defines both Mmd and Mad, but the table appears to list 'Min dist obs' and 'Avg dis obs'. Also, the sentence comparing FCP with the baselines on 'Mmd and Mmd' should read 'Mmd and Mad.'
  4. [References] References [13] and [33] are the same paper (Jordan and Perez, 'Optimal bidirectional rapidly-exploring random trees', 2013). Please consolidate or remove the duplicate.
  5. [III-D.3, Remark 5] Remark 5 states that the optimization problem has 'no inequality constraints,' but Eqs. (28) and (30) are inequality constraints on the corridor bounds and control inputs. The remark should be worded more precisely, e.g., 'no collision-avoidance inequality constraints beyond the precomputed corridor bounds.'
  6. [III-D.1, Fig. 5 caption] In the caption of Fig. 5, the notation 'dk' should be 'δ_k' to match the variables used in the equations.

Circularity Check

0 steps flagged · score 0.0 of 10

FCP's derivation is self-contained; self-citations are modular and not load-bearing.

full rationale

No circular step is present. The core derivation starts from a standard kinematic bicycle model (Eqs. 2-5), changes the independent variable to arc length (Eqs. 6-8), and then introduces the Frenet projection Δs = Δl cos(ψ+β) to obtain the space-domain kinematics (Eqs. 9-11) and the optimization dynamics (Eqs. 25-27). The output path is not used to define the model, nor is any fitted parameter relabeled as a prediction. The corridor bounds come from obstacle geometry via the Boundary Generator, and the path is optimized against those bounds with explicit smoothness, clearance, and risk terms. The paper's use of prior work is modular: the speed planner [3], perception/localization [26], and control [22] are external modules, and Remark 1 explicitly states that any external algorithm can replace them, so these self-citations are not load-bearing for FCP's path-planning claim. Remark 4 acknowledges that the bicycle model does not translate directly to the Frenet frame and that a curvature-based correction is added; this is a modeling-accuracy caveat, not a circularity, because it does not assume the conclusion it is meant to establish. The validation compares FCP against A*, RRT*, B-RRT*, CARLA baselines, and hardware experiments, providing external benchmarks rather than self-referential confirmation.

Assumptions & free parameters 8 free parameters · 7 assumptions · 0 invented entities

No new physical entities are postulated; safety-augmented boxes, corridors, and the spatial kinematics model are mathematical constructs. The central claim depends on hand-chosen objective weights, geometric margins, clustering parameters, and decision thresholds, none of which are reported.

free parameters (8)
  • Objective weight matrices Q_d, Q_u = not reported
    Weights for route deviation and steering effort in Eq (24), chosen by hand and not reported.
  • Curvature penalty weight lambda_curve = not reported
    Penalizes tan^2 u_k in Eq (24) to limit path curvature.
  • Boundary distance penalty weight lambda_risk = not reported
    Penalizes distance from the corridor center in Eq (24).
  • Dynamic obstacle penalty weight lambda_dyn = not reported
    Weights the inverse-square lateral risk sum in Eq (31).
  • Slack weight lambda_alpha and slack cap alpha_bar = not reported
    Controls bound relaxation under perception noise in Eq (32); the paper only states lambda_alpha >> 0.
  • Planning horizon N and longitudinal step Delta_s = not reported
    Set L = N * Delta_s, the planning distance and resolution; values are not given.
  • Safety augmentation margins and DBSCAN clustering parameters = not reported
    Expand vehicle boxes and cluster pedestrians before corridor generation; exact margins and DBSCAN settings are not reported.
  • Decision governor thresholds and gap cost parameters = not reported
    The decision tree in Fig. 3 selects lower or upper deviation sides; its thresholds and cost parameters are not specified.
assumptions (7)
  • domain assumption The kinematic bicycle model with no sideslip accurately represents ego vehicle motion.
    Basis for Eqs (2)-(5) and all subsequent model derivations; model error under aggressive maneuvers is not validated.
  • standard math For a in [0,1], cos^2 delta + a^2 sin^2 delta <= 1 for all delta.
    Invoked in the proof that beta approximately a*delta under-approximates |beta| in Section III-D.1; the inequality is elementary and correct.
  • domain assumption Reference path curvature can be accounted for by restricting steering bounds rather than correcting the Frenet dynamics.
    Section III-D.2 imposes curvature-induced actuation limits, while the dropped (1 - kappa*d) coupling in the Frenet transport equation is not included in the model.
  • domain assumption The decision governor correctly classifies each static obstacle as lower or upper corridor bound.
    A wrong deviation side can make the corridor infeasible or unsafe; the paper notes inconsistent classification under large perception noise.
  • domain assumption Dynamic obstacle safety in space-time is guaranteed by the downstream speed planner.
    Section III-D.4 delegates space-time collision avoidance to the speed planner from prior work; the path planner only adds a risk penalty.
  • domain assumption Bounded perception noise can always be handled by the slack variable with cap alpha_bar.
    Eq (32) relaxes the corridor bounds, but no feasibility condition is proven if noise exceeds the cap or if upper and lower bounds intersect.
  • domain assumption Safety-augmented bounding boxes and DBSCAN convex hulls conservatively cover obstacle geometry.
    The Data Processor in Section III-A assumes these representations contain the true obstacle extents; conservativeness is not proven.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Frenet Corridor Planner: An Optimal Local Path Planning Framework for Autonomous Driving." pith.science (2026). https://pith.science/paper/DCXV24LI

@misc{pith2026250503695,
  author       = {Pith},
  title        = {Pith review of: Frenet Corridor Planner: An Optimal Local Path Planning Framework for Autonomous Driving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/DCXV24LI}},
  note         = {Machine review of arXiv:2505.03695}
}
read the original abstract

Motivated by the requirements for effectiveness and efficiency, path-speed decomposition-based trajectory planning methods have widely been adopted for autonomous driving applications. While a global route can be pre-computed offline, real-time generation of adaptive local paths remains crucial. Therefore, we present the Frenet Corridor Planner (FCP), an optimization-based local path planning strategy for autonomous driving that ensures smooth and safe navigation around obstacles. Modeling the vehicles as safety-augmented bounding boxes and pedestrians as convex hulls in the Frenet space, our approach defines a drivable corridor by determining the appropriate deviation side for static obstacles. Thereafter, a modified space-domain bicycle kinematics model enables path optimization for smoothness, boundary clearance, and dynamic obstacle risk minimization. The optimized path is then passed to a speed planner to generate the final trajectory. We validate FCP through extensive simulations and real-world hardware experiments, demonstrating its efficiency and effectiveness.

Figures

Figures reproduced from arXiv: 2505.03695 by the authors.

Figure 1
Figure 1. Motivational scenario. The ego vehicle (in green) must deviate from the lane center to avoid a collision with the parked cars on the roadside while being cognizant of the oncoming traffic. Without a local path planner, the ego vehicle may remain stuck, waiting indefinitely for the parked cars to move before proceeding along its pre-determined global route. Formal optimization methods can directly incorporate dynamic… view at source ↗
Figure 2
Figure 2. Trajectory planning pipeline. The data flow between the various building blocks of FCP is illustrated on the left, while the output visualization from each module is shown on the right. algorithm. Any pedestrian not associated with a cluster is treated as an independent obstacle. At the current time t, the obstacle set containing the linearly interpolated points along the edges of the bounding boxes and convex hulls… view at source ↗
Figure 3
Figure 3. Decision tree for boundary classification of each obstacle. The decision tree evaluates the lower and upper gaps within the drivable space. If both gaps are available, two approaches can be used: selecting the preferred gap based on cost evaluation or treating the obstacle as a risk in PO [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Boundary Generation. With the pedestrians shown as blue dots, the augmented vehicle boundary depicted by green dots, and the convex hulls of pedestrian clusters given by the blue lines, Algorithm 1 generates the lower and upper bounds, shown by the red and blue lines, …
Figure 5
Figure 5. Figure 5: Numerical validation for lr lf +lr |δk| under-approximating |βk|. The linear plot lr lf +lr δk stays below the βk curve for dk ∈ [0, π 2 ) and above βk for dk ∈ (− π 2 , 0] showing |βk| ≥ lr lf +lr |δk| ∀δk ∈ (− π 2 , π 2 ). This approximation is kinematically valid (f…
Figure 6
Figure 6. Figure 6: Testing Scenario for Comparative Analysis. The ego vehicle is depicted in blue, the oncoming vehicle in red, the stationary vehicles in gray, their bounding boxes with red circles, the upper/lower bound in dashed blue/red lines, the noisy perception/prediction in light…
Figure 8
Figure 8. Figure 8: CARLA Simulations. The numbered frames show the progression of the ego vehicle through the scenario. The Rviz windows, below the Carla Pygame windows, show various objects considered during planning and the output of FCP. The ego vehicle is depicted in blue, and the ob…
Figure 9
Figure 9. Figure 9: Computational Efficiency Analysis. The test scenario in [PITH_FULL_IMAGE:figures/full_fig_p007_9.png]
Figure 10
Figure 10. Figure 10: Hardware Demonstration. A scenario involving two parked vehi￾cles and an oncoming vehicle is demonstrated using 1/10-scale autonomous cars. The numbered frames illustrate the progression of the scenario, with each frame containing snapshots of the robots on the test t…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 18 canonical work pages

  1. [3]

    Multi-profile quadratic programming (mpqp) for optimal gap selection and speed planning of autonomous driving,

    A. M. A ˜non, S. Bae, M. Saroya, and D. Isele, “Multi-profile quadratic programming (mpqp) for optimal gap selection and speed planning of autonomous driving,” arXiv preprint arXiv:2401.06305 , 2024

  2. [1]

    Toward efficient trajectory planning: The path-velocity decomposition,

    K. Kant and S. W. Zucker, “Toward efficient trajectory planning: The path-velocity decomposition,” The international journal of robotics research, vol. 5, no. 3, pp. 72–89, 1986

  3. [2]

    Dynamic trajectory planning, path- velocity decomposition and adjacent paths,

    T. Fraichard and C. Laugier, “Dynamic trajectory planning, path- velocity decomposition and adjacent paths,” in IJCAI, 1993, pp. 1592– 1599

  4. [4]

    D* lite,

    S. Koenig and M. Likhachev, “D* lite,” in Eighteenth national conference on Artificial intelligence , 2002, pp. 476–483

  5. [5]

    Anytime dynamic a*: An anytime, replanning algorithm

    M. Likhachev, D. I. Ferguson, G. J. Gordon, A. Stentz, and S. Thrun, “Anytime dynamic a*: An anytime, replanning algorithm.” in ICAPS, vol. 5, 2005, pp. 262–271

  6. [6]

    Delayed-Decision Motion Planning in the Presence of Multiple Predictions

    D. Isele, A. M. Anon, F. M. Tariq, G. Yeh, A. Singh, and S. Bae, “Delayed-decision motion planning in the presence of multiple pre- dictions,” arXiv preprint arXiv:2502.20636 , 2025

  7. [7]

    Neural informed rrt* with point-based network guidance for optimal sampling-based path planning,

    Z. Huang, H. Chen, and K. Driggs-Campbell, “Neural informed rrt* with point-based network guidance for optimal sampling-based path planning,” arXiv preprint arXiv:2309.14595 , 2023

  8. [8]

    End-to-end predictive planner for autonomous driving with consistency models,

    A. Li, S. Bae, D. Isele, R. Beeson, and F. M. Tariq, “End-to-end predictive planner for autonomous driving with consistency models,” arXiv preprint arXiv:2502.08033 , 2025

Show all 33 references
  1. [9]

    A potential field approach to path planning

    Y . K. Hwang, N. Ahuja et al. , “A potential field approach to path planning.” IEEE transactions on robotics and automation, vol. 8, no. 1, pp. 23–32, 1992

  2. [10]

    A formal basis for the heuristic determination of minimum cost paths,

    P. E. Hart, N. J. Nilsson, and B. Raphael, “A formal basis for the heuristic determination of minimum cost paths,” IEEE transactions on Systems Science and Cybernetics , vol. 4, no. 2, pp. 100–107, 1968

  3. [11]

    Theta*: Any-angle path planning on grids,

    K. Daniel, A. Nash, S. Koenig, and A. Felner, “Theta*: Any-angle path planning on grids,” Journal of Artificial Intelligence Research , vol. 39, pp. 533–579, 2010

  4. [12]

    Incremental sampling-based algorithms for optimal motion planning,

    S. Karaman and E. Frazzoli, “Incremental sampling-based algorithms for optimal motion planning,” 2011

  5. [14]

    A real-time motion planner with trajectory optimization for autonomous vehicles,

    W. Xu, J. Wei, J. M. Dolan, H. Zhao, and H. Zha, “A real-time motion planner with trajectory optimization for autonomous vehicles,” in 2012 IEEE International Conference on Robotics and Automation . IEEE, 2012, pp. 2061–2067

  6. [15]

    Optimal and reac- tive control for agile drone flight in cluttered environments,

    D. Dirckx, M. Bos, W. Decr ´e, and J. Swevers, “Optimal and reac- tive control for agile drone flight in cluttered environments,” IFAC- PapersOnLine, vol. 56, no. 2, pp. 6273–6278, 2023

  7. [17]

    Rcms: Risk-aware crash mitigation system for autonomous vehicles,

    F. M. Tariq, D. Isele, J. S. Baras, and S. Bae, “Rcms: Risk-aware crash mitigation system for autonomous vehicles,” in 2023 IEEE 26th International Conference on Intelligent Transportation Systems (ITSC). IEEE, 2023, pp. 3950–3957

  8. [18]

    Covernet: Multimodal behavior prediction using trajectory sets,

    T. Phan-Minh, E. C. Grigore, F. A. Boulton, O. Beijbom, and E. M. Wolff, “Covernet: Multimodal behavior prediction using trajectory sets,” in Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2020, pp. 14 074–14 083

  9. [19]

    Tra- jectron++: Dynamically-feasible trajectory forecasting with heteroge- neous data,

    T. Salzmann, B. Ivanovic, P. Chakravarty, and M. Pavone, “Tra- jectron++: Dynamically-feasible trajectory forecasting with heteroge- neous data,” in Computer Vision–ECCV 2020: 16th European Confer- ence, Glasgow, UK, August 23–28, 2020, Proceedings, Part XVIII 16 . Springer, 2...

  10. [20]

    Planning for autonomous cars that leverage effects on human actions

    D. Sadigh, S. Sastry, S. A. Seshia, and A. D. Dragan, “Planning for autonomous cars that leverage effects on human actions.” in Robotics: Science and systems , vol. 2. Ann Arbor, MI, USA, 2016, pp. 1–9

  11. [21]

    Re- inforcement learning for autonomous driving with latent state infer- ence and spatial-temporal relationships,

    X. Ma, J. Li, M. J. Kochenderfer, D. Isele, and K. Fujimura, “Re- inforcement learning for autonomous driving with latent state infer- ence and spatial-temporal relationships,” in 2021 IEEE International Conference on Robotics and Automation (ICRA) . IEEE, 2021, pp. 6064–6071

  12. [22]

    Lane-change in dense traffic with model predictive control and neural networks,

    S. Bae, D. Isele, A. Nakhaei, P. Xu, A. M. A ˜non, C. Choi, K. Fujimura, and S. Moura, “Lane-change in dense traffic with model predictive control and neural networks,” IEEE Transactions on Control Systems Technology, vol. 31, no. 2, pp. 646–659, 2022

  13. [23]

    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 , p. 02783649231215371, 2023

  14. [24]

    Au- tonomous vehicle overtaking in a bidirectional mixed-traffic setting,

    F. M. Tariq, N. Suriyarachchi, C. Mavridis, and J. S. Baras, “Au- tonomous vehicle overtaking in a bidirectional mixed-traffic setting,” in 2022 American Control Conference (ACC). IEEE, 2022, pp. 3132– 3139

  15. [25]

    Sur les courbes `a double courbure,

    F. Frenet, “Sur les courbes `a double courbure,” Journal de math´ematiques pures et appliqu ´ees, vol. 17, pp. 437–447, 1852

  16. [26]

    Slas: Speed and lane advisory system for highway navigation,

    F. M. Tariq, D. Isele, J. S. Baras, and S. Bae, “Slas: Speed and lane advisory system for highway navigation,” in 2022 IEEE 61st Conference on Decision and Control (CDC) . IEEE, 2022, pp. 6979– 6986

  17. [27]

    Carla: An open urban driving simulator,

    A. Dosovitskiy, G. Ros, F. Codevilla, A. Lopez, and V . Koltun, “Carla: An open urban driving simulator,” in Conference on robot learning . PMLR, 2017, pp. 1–16

  18. [28]

    Vehicle steering control with mpc for target trajectory tracking of autonomous reverse parking,

    T. Tashiro, “Vehicle steering control with mpc for target trajectory tracking of autonomous reverse parking,” in 2013 ieee international conference on control applications (cca) . IEEE, 2013, pp. 247–251

  19. [29]

    A density-based algorithm for discovering clusters in large spatial databases with noise,

    M. Ester, H.-P. Kriegel, J. Sander, X. Xu et al. , “A density-based algorithm for discovering clusters in large spatial databases with noise,” in kdd, vol. 96, no. 34, 1996, pp. 226–231

  20. [30]

    Decision tree methods: applications for classification and prediction,

    Y .-Y . Song and L. Ying, “Decision tree methods: applications for classification and prediction,” Shanghai archives of psychiatry, vol. 27, no. 2, p. 130, 2015

  21. [31]

    Recent advances in motion and behavior planning techniques for software architecture of autonomous vehicles: A state-of-the-art survey,

    O. Sharma, N. C. Sahoo, and N. B. Puhan, “Recent advances in motion and behavior planning techniques for software architecture of autonomous vehicles: A state-of-the-art survey,” Engineering applica- tions of artificial intelligence , vol. 101, p. 104211, 2021

  22. [32]

    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 2015 IEEE intelligent vehicles symposium (IV) . IEEE, 2015, pp. 1094–1099

  23. [33]

    Optimal bidirectional rapidly-exploring random trees,

    M. Jordan and A. Perez, “Optimal bidirectional rapidly-exploring random trees,” 2013

  24. [34]

    Large-scale nonlinear programming using ipopt: An integrating framework for enterprise-wide dynamic optimization,

    L. T. Biegler and V . M. Zavala, “Large-scale nonlinear programming using ipopt: An integrating framework for enterprise-wide dynamic optimization,” Computers & Chemical Engineering , vol. 33, no. 3, pp. 575–582, 2009

  25. [35]

    Mushr: A low-cost, open-source robotic racecar for education and research,

    S. S. Srinivasa et al., “Mushr: A low-cost, open-source robotic racecar for education and research,” arXiv preprint arXiv:1908.08031 , 2019

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.