Pith. sign in

REVIEW 3 major objections 3 minor 39 references

This paper claims that planning on a problem whose free space and goal region are eroded by a conformally calibrated worst-case deviation radius turns any nominal-model planner into a chance-constrained planner with a 1-alpha task-completio

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

2026-08-01 04:48 UTC pith:NRL72CH7

load-bearing objection The tightening idea is promising and the experiments are honest, but the finite-subset score breaks the conformal guarantee; the paper overclaims Prop 3.4 as written. the 3 major comments →

arxiv 2607.22409 v1 pith:NRL72CH7 submitted 2026-07-24 cs.RO cs.SYeess.SY

Conformal Constraint Tightening for Chance-Constrained Motion Planning with Unknown Dynamics

Shubham Natraj , Bruno Sinopoli , Yiannis Kantaros This is my paper
classification cs.RO cs.SYeess.SY
keywords conformal predictionchance-constrained motion planningmodel mismatchunknown dynamicsnominal modelconstraint tighteningreach-avoidplanner-agnostic guarantee
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.

Motion planning guarantees normally hold only for the simulator model, not the real robot. This paper claims that if you first calibrate a single number—a conformal radius that bounds, with probability 1-alpha, how far the robot's true trajectory can stray from its nominal trajectory for any planning problem drawn from the same distribution—then any planner that solves a shrunken version of the problem, with obstacles inflated and the goal contracted by that radius, will produce plans that satisfy the original reach-avoid task on the true system with probability at least 1-alpha. The method is planner-agnostic: the tightened problem can be handed to any planner that can solve it under the nominal model. The result matters because it converts an intractable chance-constrained problem with unknown dynamics into a standard nominal planning problem without assuming a parametric uncertainty model. The price is conservativeness: the larger the prescribed probability, the more the free space shrinks and the harder the tightened problem becomes.

Core claim

For systems with unknown dynamics f but a nominal model f-hat, the paper defines the nonconformity score of a planning problem as the worst-case deviation, over all nominally feasible control sequences, of the true trajectory from the nominal trajectory. After calibrating this score on i.i.d. planning problems, a conformal quantile q-hat_alpha is computed such that, with marginal probability 1-alpha, the true trajectory of every nominally feasible control sequence for a fresh problem stays within Euclidean balls of radius q-hat_alpha centered on its nominal states. The paper then erodes the safe set to {x : B(x, q-hat_alpha) subset of X_free} and the goal similarly, and proves Proposition 3.

What carries the argument

The central object is the conformal threshold q-hat_alpha computed from the nonconformity score (5), the supremum over nominally feasible control sequences of the maximum per-timestep Euclidean distance between true and nominal states. Conformal prediction converts calibration scores into a finite-sample, distribution-free quantile guarantee; the paper uses that quantile to shrink the free space and goal region, turning the chance constraint (3) into a deterministic, nominal-only planning problem.

Load-bearing premise

The guarantee rests on the assumption that the conformal quantile computed from a finite subset of calibration control sequences bounds the deviation of the single control sequence returned by the test-time planner; if that planner's output deviates more than the calibration scores, the coverage guarantee (6) does not apply.

What would settle it

Run the calibrated pipeline with a test-time planner different from the one used to generate the calibration sequences (e.g., an optimization-based planner that aggressively cuts corners), and record the empirical coverage of the returned plans. If the fraction of plans whose true trajectories stay within the conformal balls falls below 1-alpha on 100 fresh problems, the central claim is falsified.

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

If this is right

  • If a planner can solve the eroded problem M_alpha for the nominal model, its output is guaranteed to satisfy the original reach-avoid specification on the true system with probability at least 1-alpha (Prop. 3.4).
  • The guarantee is marginal over the distribution of planning problems D_M, not per-task, so it applies to repeated deployments in novel environments drawn from the same distribution.
  • Increasing the required coverage 1-alpha yields a larger conformal radius, which shrinks the feasible set and can reduce planner success rate, revealing a trade-off between conservativeness and solvability.
  • The framework extends to stochastic true dynamics with bounded disturbances by replacing the score with a worst-case over disturbance realizations (Remark 3.3), at the cost of added conservativeness.
  • Calibration data can come from executing only a finite subset of control sequences per problem (Remark 3.2), making the procedure practical with modest data collection.

Where Pith is reading between the lines

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

  • The finite-subset approximation in Remark 3.2 means the guarantee (6) is only as strong as the representativeness of the calibration control sequences; a test-time planner that systematically produces control sequences with larger deviations than those in the calibration subset could invalidate the coverage. This is my own inference, not the paper's claim.
  • Because the guarantee is marginal over D_M, it does not bound the failure probability of any single mission; one could in principle combine repeated calls to the planner with adaptive risk allocation to bound per-mission risk, an extension the paper does not make.
  • The method's reliance on a single worst-case radius q-hat_alpha is conservative for states that are far from obstacles; a state-dependent or time-dependent radius, calibrated with weighted conformal prediction, could shrink the erosion and improve planner success without sacrificing the marginal guarantee.
  • If the same robot collects calibration data across different environments over time, the i.i.d. assumption on planning problems may be violated; extending to non-exchangeable data (e.g., weighted or online conformal) would make the framework applicable to lifelong deployments.

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 / 3 minor

Summary. This paper proposes a conformal-prediction-based constraint tightening method for chance-constrained motion planning when the true dynamics are unknown and only a nominal model is available. A nonconformity score is defined as the worst-case deviation, over all nominally feasible control sequences, between the nominal trajectory and the true trajectory for a planning problem; split conformal prediction is used to estimate a threshold q_hat_alpha. The free space and goal region are eroded by q_hat_alpha, and the paper claims that any control sequence solving the tightened problem under the nominal model solves the original reach-avoid problem on the true system with marginal probability at least 1-alpha. The method is evaluated on a Dubins vehicle with parametric mismatch and on a stochastic planar quadrotor with a learned nominal model.

Significance. The problem is well motivated and the idea of applying conformal prediction at the level of whole planning problems, rather than per-state prediction errors, is novel and potentially useful. The paper is clearly written, the setup is clean, and the experiments cover two realistic forms of model mismatch. However, the central theoretical guarantee is not established: the calibration uses a finite-subset approximation of the nonconformity score, while the theorem is stated for the true supremum, and no argument closes that gap. Because the chance-constrained guarantee is the main contribution, the current result cannot be recommended for publication.

major comments (3)
  1. [Section III-B, Eqs. (5)-(6); Remark 3.2; Section IV-A] The conformal bound (6) is not justified by the calibration procedure. Eq. (5) defines R(M) as a supremum over all nominally feasible control sequences U(M), but during calibration the scores are computed as tilde_R_i = max over the 10 RRT-generated sequences, not over U(M_i). Since tilde_R_i <= R_i, the empirical (1-alpha)-quantile q_hat_alpha of the approximate scores is generally smaller than the quantile of the true scores. The test-time quantity in (6) is the true supremum R(M_test), while the calibration scores are maxima over 10 random sequences; these are not exchangeable, so the standard split-conformal guarantee (4) does not apply to (6). No bound on the finite-subset approximation error is given. Thus Proposition 3.1 and Proposition 3.4, which both rely on (6), are unsupported.
  2. [Section IV-B, Empirical Coverage] The reported empirical coverage does not validate (6). It is the fraction of returned RRT plans whose executed trajectory stays within q_hat_alpha; it is not the coverage frequency of the true supremal score R(M) defined in (5). A test-time planner that returns a plan with deviation below q_hat_alpha will be counted as covered even if other nominally feasible plans have deviation above q_hat_alpha. Thus the experiments can be consistent with the proposed method while the claimed worst-case guarantee fails. To validate the theorem, the scores for the test problems would need to be evaluated over the full set U(M) or an independent upper bound.
  3. [Remark 3.3 and Section IV-A] The stochastic extension inherits the same gap in a stronger form. R_stoch(M) is defined as a supremum over control sequences and disturbance realizations, but calibration evaluates only a finite number of noisy executions of 10 RRT sequences. The quadrotor results therefore demonstrate coverage of the particular realized executions, not of the worst-case stochastic quantity in Remark 3.3. A separate argument, not present in the paper, is needed to connect the finite noisy samples to the supremum over W^T.
minor comments (3)
  1. [Section IV-A, Eq. (9)] The notation is inconsistent: both v_y,t and vy,t are used in the same equation. Please unify.
  2. [Proposition 3.4] The phrase 'with probability at least 1-alpha' should explicitly state that the probability is over the calibration data and the new problem M ~ D_M. As written, it can be read as a per-trajectory guarantee, which is not what the proof establishes.
  3. [Figure 2 caption] The caption says 'Dashed horizontal lines' for the nominal RRT baseline. If these lines are not constant across the x-axis, please clarify what is plotted.

Circularity Check

0 steps flagged

No circularity: the conformal guarantee is estimated from independent calibration data; the finite-subset NCS approximation is an admitted soundness gap, not a by-construction reduction.

full rationale

The derivation chain is not circular. The threshold qhat_alpha in (6) comes from split-conformal calibration on i.i.d. planning problems (Eq. (4) and Section III-B), not from the test outcome. Proposition 3.4 is a direct consequence of the definitions of the eroded sets in (8), the CP coverage event in (6), and Proposition 3.1; no parameter is fitted to the planner's returned trajectory. The only true limitation is the practical NCS approximation: Remark 3.2 explicitly says 'Computing the NCS in (5) exactly requires optimizing over all nominally feasible control sequences in U(M), which is generally intractable' and that it is approximated 'using a finite set of control sequences'; Section IV-A implements this with 10 RRT sequences. This makes the calibration scores maxima over finite subsets rather than the supremum in (5), so the empirical quantile need not bound the true supremal test score and the coverage statement (6) is not rigorously delivered as implemented. That is a soundness/coverage gap, not a circular step: the formal guarantee is stated for the exact score, and the approximation is openly acknowledged rather than hidden in a definition. The one self-citation, [36] in Remark 3.2, supports the approximation but is cited together with external [37] and is not load-bearing for the main theorem. Accordingly, no load-bearing step reduces to its inputs, and the paper is in the 0-2 band; the score reflects only the minor non-load-bearing self-citation.

Axiom & Free-Parameter Ledger

2 free parameters · 3 axioms · 0 invented entities

The method relies on standard conformal-prediction exchangeability, the accessibility of the true system for calibration, and — critically — the unverified assumption that a finite set of 10 RRT plans suffices to compute a conservative deviation bound. No new physical entities are introduced. The free parameters are experiment choices, not fitted constants, but the finite-subset size is load-bearing for the guarantee.

free parameters (2)
  • Finite-subset size for NCS approximation (10 RRT plans) = 10
    The nonconformity score is approximated with 10 RRT-generated control sequences per calibration problem (Remark 3.2). This hand-chosen number affects the validity of the conformal threshold and is central to the proof gap.
  • Calibration dataset size Ncal = 100
    The number of calibration planning problems is set to 100. It is chosen by hand and affects the variance of the conformal threshold, though the coverage guarantee formally holds for any Ncal under exchangeability.
axioms (3)
  • domain assumption Planning problems M are drawn i.i.d. from an unknown distribution D_M, and calibration problems are independent and identically distributed with future problems.
    Section II: 'We assume that planning problems are drawn independently from an unknown distribution D_M, from which i.i.d. samples can be obtained.' Required for conformal validity.
  • domain assumption The true system is physically accessible for calibration, allowing execution of nominally feasible control sequences and measurement of true trajectories.
    Section III-B: 'During calibration, we assume physical access to the true system (1), allowing nominally feasible control sequences to be executed and their resulting true trajectories collected.'
  • ad hoc to paper The finite-subset approximation of the nonconformity score (using 10 RRT plans) is representative enough that the conformal threshold computed from it bounds the true deviation of any nominally feasible control sequence, including the test-time planner output.
    Remark 3.2 approximates the intractable supremum in (5) with a finite set. The subsequent propositions treat this approximation as exact, which is not justified and is load-bearing for Proposition 3.4.

pith-pipeline@v1.3.0-alltime-deepseek · 10617 in / 13530 out tokens · 148418 ms · 2026-08-01T04:48:59.260171+00:00 · methodology

0 comments
read the original abstract

Motion planning algorithms compute control sequences that drive autonomous robots to goal regions while avoiding unsafe states. Existing methods, from sampling-based planning to deep reinforcement learning, typically provide task-completion guarantees only with respect to a nominal model or simulator, which may be invalidated when the true dynamics are unknown or difficult to model accurately. This letter addresses this limitation for systems with unknown dynamics and an available approximate nominal model, contributing a planner-agnostic constraint-tightening procedure that equips existing planners with a probabilistic task-completion guarantee on the true system. We leverage conformal prediction to provide a probabilistic bound on the nominal-to-true trajectory deviation over a distribution of planning problems. We tighten the planning constraints using that bound, and show that solving the tightened problem under the nominal model is a sufficient condition for solving the original problem on the true system with a prescribed probability. We validate the theoretical guarantees empirically and demonstrate substantially improved task completion relative to nominal-model planning.

Figures

Figures reproduced from arXiv: 2607.22409 by Bruno Sinopoli, Shubham Natraj, Yiannis Kantaros.

Figure 1
Figure 1. Figure 1: Illustration of the proposed CP-based planning frame [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Empirical results for the Dubins vehicle (left column) [PITH_FULL_IMAGE:figures/full_fig_p005_2.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

39 extracted references · 5 linked inside Pith

  1. [1]

    Randomized kinodynamic plan- ning,

    S. M. LaValle and J. J. Kuffner Jr, “Randomized kinodynamic plan- ning,”The international journal of robotics research, vol. 20, no. 5, pp. 378–400, 2001

    2001

  2. [2]

    Asymptotically optimal sampling-based kinodynamic planning,

    Y . Li, Z. Littlefield, and K. E. Bekris, “Asymptotically optimal sampling-based kinodynamic planning,”The International Journal of Robotics Research, vol. 35, no. 5, pp. 528–564, 2016

    2016

  3. [3]

    An abstraction-free method for multirobot temporal logic optimal control synthesis,

    X. Luo, Y . Kantaros, and M. M. Zavlanos, “An abstraction-free method for multirobot temporal logic optimal control synthesis,”IEEE Transactions on Robotics, vol. 37, no. 5, pp. 1487–1507, 2021

    2021

  4. [4]

    Kino-pax: Highly parallel kinodynamic sampling-based planner,

    N. Perrault, Q. H. Ho, and M. Lahijanian, “Kino-pax: Highly parallel kinodynamic sampling-based planner,”IEEE Robotics and Automation Letters, vol. 10, no. 3, pp. 2430–2437, 2025

    2025

  5. [5]

    Ultrafast sampling-based kinodynamic planning via differ- ential flatness,

    T. Duong, C. W. Ramsey, Z. Kingston, W. Thomason, and L. E. Kavraki, “Ultrafast sampling-based kinodynamic planning via differ- ential flatness,”arXiv preprint arXiv:2603.16059, 2026

    Pith/arXiv arXiv 2026

  6. [6]

    A rapidly-exploring random trees motion planning algorithm for hybrid dynamical systems,

    N. Wang and R. G. Sanfelice, “A rapidly-exploring random trees motion planning algorithm for hybrid dynamical systems,” in2022 IEEE 61st Conference on Decision and Control (CDC), 2022, pp. 2626–2631

    2022

  7. [7]

    Model predictive control with preview: Recursive feasibility and stability,

    X. Fang and W.-H. Chen, “Model predictive control with preview: Recursive feasibility and stability,”IEEE Control Systems Letters, vol. 6, pp. 2647–2652, 2022

    2022

  8. [8]

    Online trajectory generation with distributed model predictive control for multi-robot motion planning,

    C. E. Luis, M. Vukosavljev, and A. P. Schoellig, “Online trajectory generation with distributed model predictive control for multi-robot motion planning,”IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 604–611, 2020

    2020

  9. [9]

    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

  10. [10]

    Deepgait: Planning and control of quadrupedal gaits using deep reinforcement learning,

    V . Tsounis, M. Alge, J. Lee, F. Farshidian, and M. Hutter, “Deepgait: Planning and control of quadrupedal gaits using deep reinforcement learning,”IEEE Robotics and Automation Letters, vol. 5, no. 2, pp. 3699–3706, 2020

    2020

  11. [11]

    Semantically-driven deep reinforcement learning for inspection path planning,

    G. Malczyk, M. Kulkarni, and K. Alexis, “Semantically-driven deep reinforcement learning for inspection path planning,”IEEE Robotics and Automation Letters, 2025

    2025

  12. [12]

    Safe multiagent motion planning under uncertainty for drones using filtered reinforcement learning,

    S. Safaoui, A. P. Vinod, A. Chakrabarty, R. Quirynen, N. Yoshikawa, and S. Di Cairano, “Safe multiagent motion planning under uncertainty for drones using filtered reinforcement learning,”IEEE Transactions on Robotics, vol. 40, pp. 2529–2542, 2024

    2024

  13. [13]

    Learning-based near-optimal motion planning for intelligent vehicles with uncertain dynamics,

    Y . Lu, X. Zhang, X. Xu, and W. Yao, “Learning-based near-optimal motion planning for intelligent vehicles with uncertain dynamics,” IEEE Robotics and Automation Letters, vol. 9, no. 2, pp. 1532–1539, 2023

    2023

  14. [14]

    Optimal robot motion planning in constrained workspaces using reinforcement learning,

    P. Rousseas, C. P. Bechlioulis, and K. J. Kyriakopoulos, “Optimal robot motion planning in constrained workspaces using reinforcement learning,” in2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2020, pp. 6917–6922

    2020

  15. [15]

    Chance constrained rrt for prob- abilistic robustness to environmental uncertainty,

    B. Luders, M. Kothari, and J. How, “Chance constrained rrt for prob- abilistic robustness to environmental uncertainty,” inAIAA guidance, navigation, and control conference, 2010, p. 8160

    2010

  16. [16]

    Distributionally robust sampling-based motion planning under uncertainty,

    T. Summers, “Distributionally robust sampling-based motion planning under uncertainty,” in2018 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2018, pp. 6518–6523

    2018

  17. [17]

    Provably safe motion planning under unknown disturbances,

    I. Gracia, Q. H. Ho, L. Laurenti, and M. Lahijanian, “Provably safe motion planning under unknown disturbances,”arXiv preprint arXiv:2605.26625, 2026

    Pith/arXiv arXiv 2026

  18. [18]

    Robust- rrt: Probabilistically-complete motion planning for uncertain nonlinear systems,

    A. Wu, T. Lew, K. Solovey, E. Schmerling, and M. Pavone, “Robust- rrt: Probabilistically-complete motion planning for uncertain nonlinear systems,” inThe International Symposium of Robotics Research. Springer, 2022, pp. 538–554

    2022

  19. [19]

    Scenario-based stochastic mpc for systems with uncertain dynamics,

    F. Micheli and J. Lygeros, “Scenario-based stochastic mpc for systems with uncertain dynamics,” in2022 European Control Conference (ECC). IEEE, 2022, pp. 833–838

    2022

  20. [20]

    Data-driven distributionally robust mpc for systems with uncertain dynamics,

    F. Micheli, T. Summers, and J. Lygeros, “Data-driven distributionally robust mpc for systems with uncertain dynamics,” inIEEE 61st Conference on Decision and Control (CDC), 2022, pp. 4788–4793

    2022

  21. [21]

    Can’t touch this: Real-time, safe motion planning and control for manipulators under uncertainty,

    J. Michaux, P. Holmes, B. Zhang, C. Chen, B. Wang, S. Sahgal, T. Zhang, S. Dey, S. Kousik, and R. Vasudevan, “Can’t touch this: Real-time, safe motion planning and control for manipulators under uncertainty,”IEEE Transactions on Robotics, 2025

    2025

  22. [22]

    Spatiotemporal tubes for differential drive robots with model uncertainty,

    R. Das, A. Basu, C. Verginis, and P. Jagtap, “Spatiotemporal tubes for differential drive robots with model uncertainty,”arXiv preprint arXiv:2512.05495, 2025

    Pith/arXiv arXiv 2025

  23. [23]

    Tube-based robust nonlinear model predictive control,

    D. Q. Mayne, E. C. Kerrigan, E. Van Wyk, and P. Falugi, “Tube-based robust nonlinear model predictive control,”International journal of robust and nonlinear control, vol. 21, no. 11, pp. 1341–1353, 2011

    2011

  24. [24]

    Distributionally robust uncertainty quan- tification via data-driven stochastic optimal control,

    G. Pan and T. Faulwasser, “Distributionally robust uncertainty quan- tification via data-driven stochastic optimal control,”IEEE Control Systems Letters, vol. 7, pp. 3036–3041, 2023

    2023

  25. [25]

    Kdf: Kin- odynamic motion planning via geometric sampling-based algorithms and funnel control,

    C. K. Verginis, D. V . Dimarogonas, and L. E. Kavraki, “Kdf: Kin- odynamic motion planning via geometric sampling-based algorithms and funnel control,”IEEE Transactions on robotics, vol. 39, no. 2, pp. 978–997, 2022

    2022

  26. [26]

    Conformal prediction for distribution-free optimal control of linear stochastic systems,

    E. E. Vlahakis, L. Lindemann, P. Sopasakis, and D. V . Dimarogonas, “Conformal prediction for distribution-free optimal control of linear stochastic systems,”IEEE Control Systems Letters, vol. 8, pp. 2835– 2840, 2024

    2024

  27. [27]

    Uncertainty quantification for learning-based mpc using weighted conformal prediction,

    K. Y . Chee, M. A. Hsieh, and G. J. Pappas, “Uncertainty quantification for learning-based mpc using weighted conformal prediction,” in62nd IEEE Conference on Decision and Control (CDC), 2023, pp. 342–349

    2023

  28. [28]

    Uncertainty quantification and robustification of model-based controllers using conformal prediction,

    K. Y . Chee, T. C. Silva, M. A. Hsieh, and G. J. Pappas, “Uncertainty quantification and robustification of model-based controllers using conformal prediction,” in6th Annual Learning for Dynamics & Control Conference. PMLR, 2024, pp. 528–540

    2024

  29. [29]

    Safety beyond the training data: Robust out-of-distribution mpc via conformalized system level synthesis,

    A. Srinivasan, A. Leeman, and G. Chou, “Safety beyond the training data: Robust out-of-distribution mpc via conformalized system level synthesis,” inLearning for Dynamics and Control, 2026

    2026

  30. [30]

    Motion planning and control with unknown nonlinear dynamics through predicted reachability,

    Z. Zhang, G. Puthumanaillam, M. V ora, and M. Ornik, “Motion planning and control with unknown nonlinear dynamics through predicted reachability,” in2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). IEEE, 2025, pp. 342–349

    2025

  31. [31]

    Active learning of discrete-time dynamics for uncertainty-aware model pre- dictive control,

    A. Saviolo, J. Frey, A. Rathod, M. Diehl, and G. Loianno, “Active learning of discrete-time dynamics for uncertainty-aware model pre- dictive control,”IEEE Transactions on Robotics, vol. 40, pp. 1273– 1291, 2023

    2023

  32. [32]

    Residual reinforcement learning for robot control,

    T. Johannink, S. Bahl, A. Nair, J. Luo, A. Kumar, M. Loskyll, J. A. Ojea, E. Solowjow, and S. Levine, “Residual reinforcement learning for robot control,”arXiv preprint arXiv:1812.03201, 2018

    Pith/arXiv arXiv 2018

  33. [33]

    Sim-to- real transfer of robotic control with dynamics randomization,

    X. B. Peng, M. Andrychowicz, W. Zaremba, and P. Abbeel, “Sim-to- real transfer of robotic control with dynamics randomization,” in2018 IEEE international conference on robotics and automation (ICRA). IEEE, 2018, pp. 3803–3810

    2018

  34. [34]

    Conformal prediction: A gentle introduction,

    A. N. Angelopoulos and S. Bates, “Conformal prediction: A gentle introduction,”F oundations and Trends in Machine Learning, vol. 16, no. 4, pp. 494–591, 2023

    2023

  35. [35]

    A tutorial on conformal prediction

    G. Shafer and V . V ovk, “A tutorial on conformal prediction.”Journal of machine learning research, vol. 9, no. 3, 2008

    2008

  36. [36]

    Safe planning in unknown environments using confor- malized semantic maps,

    D. S. Sundarsingh, Y . Li, T. Tang, G. J. Pappas, N. Atanasov, and Y . Kantaros, “Safe planning in unknown environments using confor- malized semantic maps,”IEEE Robotics and Automation Letters, 2026

    2026

  37. [37]

    Perceive with confidence: Statistical safety assurances for navigation with learning-based perception,

    Z. Mei, A. Dixit, M. Booker, E. Zhou, M. Storey-Matsutani, A. Z. Ren, O. Shorinwa, and A. Majumdar, “Perceive with confidence: Statistical safety assurances for navigation with learning-based perception,”The International Journal of Robotics Research, vol. 45, no. 6, pp. 938– 967, 2026

    2026

  38. [38]

    Rapidly-exploring random trees: A new tool for path planning,

    S. LaValle, “Rapidly-exploring random trees: A new tool for path planning,”Research Report 9811, 1998

    1998

  39. [39]

    Adapting neural robot dynamics on the fly for predictive control,

    A. Altawaitan and N. Atanasov, “Adapting neural robot dynamics on the fly for predictive control,”arXiv preprint arXiv:2604.04039, 2026

    Pith/arXiv arXiv 2026