Pith. sign in

REVIEW 4 major objections 5 minor 26 references

Robust Optimal Safe and Stability Guaranteeing Reinforcement Learning Control for Quadcopter

T0 review · 4 major / 5 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read This paper claims that replacing the infinity-norm Lipschitz bound with an L2-norm, SDP-estimated bound makes LMI-based robust stability certificates practical for neural-network quadcopter control, certifying a safe invariant domain…

desk verdict The L2-norm reformulation is sound, but the quadcopter paper never shows the LMI-certified nominal gain K equals the simulated PD controller, so the stability guarantee does not cover the demonstrated system. read the letter →

arxiv 2412.14003 v2 pith:SENSNHHL submitted 2024-12-18 eess.SY cs.SY

Sanghyoup Gu , Ratnesh Kumar This is my paper · ORCID
classification eess.SYcs.SY
keywords neuralnetworkcontrolquadcopterLyapunovstabilityLipschitzboundreinforcementlearninglinearmatrixinequalitysafeinvariantsetrobust
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 extends the ROSS-GC method of [1] to certify a neural-network controller for a quadcopter. The central claim is that the closed-loop system, a nominal state-feedback controller plus an RL-trained neural network, is robustly asymptotically stable and safe inside a computed invariant domain $S^*$ despite random $\pm5\%$ variations in thrust and torque. The certificate is a linear matrix inequality whose feasibility yields a quadratic Lyapunov function, with the neural network's contribution bounded by a Lipschitz constant. The paper's key improvement is replacing the infinity-norm Lipschitz bound with an L2-norm formulation and estimating the constant by semidefinite programming, reducing conservativeness by more than 2400-fold for the quadcopter example. If the certificate is valid, it offers a practical route to deploying learned controllers in safety-critical flight with formal safety and stability guarantees.

What carries the argument

The load-bearing object is the LMI constraint (9). Its feasibility implies the existence of a robust quadratic Lyapunov function $V(s)=s^T P s$ and an invariant sublevel set contained in the user-specified safety domain, giving both safety and asymptotic stability under bounded parameter variation. The LMI is assembled from two quadratic constraints: one expressing the NN controller's Lipschitz bound and one capturing the sector bounds of the nonlinear-plus-parameter-variation part of the closed-loop dynamics. The enhancements are the Euclidean-norm version of the Lipschitz constraint, the SDP-based Lipschitz estimator, and a final linear layer that can be scaled down to meet the certified bound without changing the action pattern.

What would settle it

Simulate the actual cascade-PD-plus-NN closed loop from many initial states inside $S^*$ under $\pm5\%$ thrust and torque variations; if any trajectory leaves the claimed safe bounds or fails to converge, the certificate does not cover the controller as implemented.

Watch

Extended reading notes

Core claim

The paper's central claim is that a robust, safe, and asymptotically stabilizing NN controller can be synthesized for a quadcopter by keeping the NN's $L_2$-Lipschitz bound below $L^* = 1.2613$ and starting inside the invariant domain $S^*$, with position coordinates bounded by $\pm 1.892$ m, Euler-angle coordinates by $\pm 0.252$ rad, linear velocities by $\pm 1.892$ m/s, and angular velocities by $\pm 0.266$ rad/s. Feasibility of the LMI produces a robust quadratic Lyapunov function $V(s)=s^T P s$ whose sublevel sets certify asymptotic convergence despite $\pm5\%$ thrust and torque variations. The paper shows that using the Euclidean norm instead of the infinity norm preserves the LMI's feasibility while making the Lipschitz bound far less conservative, and that an SDP-based Lipschitz estimate lowers the bound from 2920 to 1.2608 for the trained actor after scaling the final linear layer by 0.4. The resulting controller therefore has a certified Lipschitz constant $1.2608 < L^*$.

Load-bearing premise

The certificate assumes the nominal controller is exactly the linear state feedback $u_0 = K s$, whereas the controller implemented and simulated is a nonlinear cascade PD controller, and the mismatch is not included in the sector bounds used to construct the LMI.

Editorial extensions

If this is right

  • The trained NN controller, with Lipschitz bound $1.2608 < L^* = 1.2613$, is certified to asymptotically stabilize the quadcopter from every initial state inside $S^*$ under random $\pm5\%$ thrust and torque variation.
  • The Euclidean-norm version of the Lipschitz quadratic constraint preserves LMI feasibility, making the ROSS-GC certificate practical for high-dimensional neural networks where the infinity-norm bound was too conservative.
  • SDP-based Lipschitz estimation reduces the estimated bound from 2920 to 1.2608 for the quadcopter example, a more-than-2400-fold reduction in conservativeness.
  • Scaling the final linear layer of the trained actor gives a certified controller while retaining the learned action pattern.

Reading between the lines

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

  • If the nominal controller in the LMI were replaced by the cascade PD controller used in the simulations, the sector bounds would change; the computed $L^* = 1.2613$ and the simulated trajectories are therefore not yet covered by the same certificate.
  • The Euclidean-norm relaxation is not specific to quadcopters and should extend to any Lipschitz-bounded learned feedback law, potentially easing certified learning-based control for other nonlinear systems.
  • The certificate applies to a fixed equilibrium; proving tracking of the time-varying S-shaped trajectory would require reformulating the LMI on tracking-error dynamics with the reference treated as an exogenous input.
  • A direct testable next step is to recompute the LMI with NPV sector bounds taken from the cascade-PD closed-loop Jacobian and compare the maximal feasible Lipschitz bound with 1.2613.
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 / 5 minor

Summary. This paper applies and extends the ROSS-GC framework of [1] to quadcopter control. The proposed modifications replace the infinity-norm Lipschitz condition with an L2-norm condition, use an SDP-based Lipschitz estimator [5], and scale the final linear layer of a PPO-trained actor to satisfy the certified Lipschitz bound. The authors report a maximal Lipschitz bound L*=1.2613 and a safe domain S* of coordinate intervals, and present simulation results for step and S-shaped trajectory tracking with ±5% thrust/torque uncertainty. The central claim is that the resulting NN-augmented cascade PD controller is certified robustly asymptotically stable and safe within S*.

Significance. If fully established, the L2-norm reformulation and SDP-based Lipschitz estimation would be a useful practical improvement over [1], and the reported >2400-fold reduction in conservatism (Table V) is striking. However, the paper does not currently deliver a valid stability certificate for the controller it simulates: the certified plant uses a generic state-feedback gain K, while the simulated controller is a specific cascade PD, and the connection between the two is never made. The reported safe domain is also not shown to be the invariant sublevel set required by Theorem 1. Because the central guarantee is the paper's main contribution, these gaps are decisive. The paper contains no code or full certificate data, so the numerical claims cannot be independently checked.

major comments (4)
  1. [Section III-C, Eq. (9); Section IV-A/B] The LMI certificate is for a nominal feedback of the exact form u0(s)=K s, with the NPV sector bounds in Section III-C computed for this K. Section IV-B calls K a 'state-feedback gain of a nominal controller' derived from the feasibility search, while the controller actually simulated in Section IV-A is the cascade PD controller with the gains in Table III. The paper never reports the K matrix used in the LMI, nor does it show that the cascade PD law, after the coordinate shift and the allocation map, equals K s on the certified domain S*. Consequently, Theorem 1 does not certify the closed-loop system whose responses are shown in Figures 5-7, and the central stability guarantee is not established for the demonstrated controller.
  2. [Section IV-B and Section IV-D] The paper equates the safe initial-state set with the box S* given by coordinate intervals, but Theorem 1 and Definition 1 guarantee invariance and convergence for a sublevel set {s: s^T P s <= c} of the quadratic Lyapunov function, not for a box. The authors do not report the sublevel set or prove that the box S* is contained in it; therefore, the statement that starting anywhere in S* is certified safe is not supported by the presented analysis.
  3. [Section III-A and Eq. (10)] The framework assumes that the equilibrium is invariant under parameter variations, but in the quadcopter model the uncertainty multiplies the total thrust and torques. At s=0 and u_NN=0, the z-acceleration in Eq. (10) is alpha_1 g, so the origin is not an equilibrium for alpha_1 != 0. The paper neither imposes pi_NN(0)=0 nor accounts for a parameter-dependent equilibrium, so the asymptotic stability conclusion to the origin does not follow from the stated assumptions.
  4. [Section IV-C and IV-D] No numerical data for K, P, Lambda, gamma_ij, or the NPV sector bounds c_ij are provided, and no code is made available; the LMI feasibility result and the claimed L*=1.2613 are therefore not reproducible, and the >2400-fold conservatism comparison cannot be audited. For a certificate-based paper, this is not a cosmetic omission.
minor comments (5)
  1. [Introduction] There are several typos, including 'tne promising use case', 'operting', and irregular spacing in 'UA Vs'; these should be corrected.
  2. [Section IV-A, Eq. (10)] The symbol u_i is defined as the total motor thrusts, but in Eq. (10) the terms u1+...+u4 appear to represent only the NN augmentation on top of the embedded nominal PD; this notational ambiguity should be disambiguated.
  3. [Section V-B] The table of timed waypoints is labeled Table V, but Table V already denotes the Lipschitz estimates in Section IV-D; the tables should be renumbered.
  4. [Section IV-C] Proximity Policy Optimization should be Proximal Policy Optimization.
  5. [Section IV-A] The allocation matrix inverse is only given implicitly; to make the nominal controller reproducible, present the explicit inverse matrix or state the resulting equations for u_i in terms of Td and tau_d.

Circularity Check

1 steps flagged · score 4.0 of 10

Central stability certificate is imported from the authors' own prior paper [1] rather than re-derived; new L2/SDP enhancements are non-circular.

  1. self citation load bearing [Section III-D, Theorem 1 (LMI constraint)]
    "Theorem 1 (Linear Matrix Inequality Constraint [1]) For the given system of (6), when the below LMI constraint of (9) is feasible for some L, K,(P ∈ S+), (Λ ≥ 0), (γij ≥ 0, ∀i ∈ 1, . . . , m, j∈ 1, . . . , n), then there exists a robust quadratic Lyapunov function V(s) = sT P ssuch that π(s) =Ks+πN N(s) with πN N∈ ΠL stabilizes the system of (6) over the given state domain under the variations of the parameters in its given domain"

    The paper's robust stability guarantee is not derived in this submission; it is quoted from [1] (Talukder and Kumar), whose second author is the present paper's coauthor. The theorem's conclusion—that feasibility of LMI (9) yields a robust quadratic Lyapunov function, an invariant safe region of stability, and asymptotic stability—is the sole load-bearing step for the central claim. The paper's new contributions (L2-norm Lipschitz continuity and SDP-based Lipschitz estimation) only relax the conservativeness of the bound L*; they do not prove or independently verify the Lyapunov implication. Thus the central certificate reduces to a self-citation that is not re-established or machine-checked here.

full rationale

The only substantive circular concern is the imported Theorem 1, which is load-bearing self-citation. The L2-norm extension is proved in-text via inequalities and the SDP Lipschitz estimator is an external method [5], so those enhancements are not circular. The final layer scaling is an explicit enforcement step, not a hidden fit, and the >2400-fold comparison, while potentially apples-to-oranges, is a presentation concern rather than a circular derivation. The suspected mismatch between the certified K in u0 = K s and the simulated cascade PD controller in Section IV-A is a real reproducibility and soundness gap, but it is not circularity: the paper never identifies the two by construction, and if the LMI-search K differs from the PD gains, the formal guarantee simply may not apply to the simulated controller. Because the central stability certificate is an unexamined import from the authors' own prior work, while the new enhancements retain independent content, the score is 4 rather than 0 or 2.

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

The central claim rests on the LMI theorem from [1], which is not independently verified in this paper, and on the assumption that the nominal controller is the linear gain K. The nonlinear cascade PD actually used is not covered by the sector bounds. The listed free parameters shape the design and the certificate but are not themselves derived from data.

free parameters (4)
  • Nominal PD controller gains = Kx=Ky=0.05, Kz=0.1, Kphi=Ktheta=0.1, Kpsi=0.1, Kdotphi=Kdottheta=0.01, Kdotpsi=0.1
    Chosen by the designer to stabilize the linearized nominal system; they determine AK and the LMI feasibility.
  • Parameter variation bounds = ±5% for thrust and torque
    Assumed bound on thrust and torque uncertainty; appears in Table II and is used in the sector bounds.
  • Reward function weights and threshold = 0.8, 0.2, 0.6, 0.2, 0.2, threshold 0.3
    Hand-picked to shape the RL objective; affects the trained policy but not the stability certificate.
  • Initial guess (L0, S0) and iterative search step sizes = L0=1.0; S0 bounds as listed
    Starting point for the maximal (L,S) search in Algorithm 1 of [1]; influences the final L* and S*.
assumptions (4)
  • domain assumption LMI (9) of [1] is a sufficient condition for robust asymptotic stability of the NN-controlled system.
    Imported from [1], not re-derived here; the paper relies on it for the entire stability guarantee.
  • domain assumption The equilibrium is invariant under parameter variations.
    Assumed to allow shifting equilibrium to origin (Section III-A).
  • ad hoc to paper The nominal controller can be represented as K s.
    Section III-C equates pi0(s) with Ks, but Section IV-A uses a cascade PD controller; this is the load-bearing gap.
  • domain assumption Parameter variations are bounded to ±5%.
    Table II; used in the sector bounds and LMI.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Robust Optimal Safe and Stability Guaranteeing Reinforcement Learning Control for Quadcopter." pith.science (2026). https://pith.science/paper/SENSNHHL

@misc{pith2026241214003,
  author       = {Pith},
  title        = {Pith review of: Robust Optimal Safe and Stability Guaranteeing Reinforcement Learning Control for Quadcopter},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SENSNHHL}},
  note         = {Machine review of arXiv:2412.14003}
}
read the original abstract

Recent advances in deep learning have provided new data-driven ways of controller design to replace the traditional manual synthesis and certification approaches. Employing neural network (NN) as controllers however, presents its own challenge: that of certifying stability due to their inherent complex nonlinearity, and while NN controllers have demonstrated high performance in complex systems, they often lack formal stability guarantees. This issue is further accentuated for critical nonlinear applications such as of unmanned aerial vehicles (UAVs), complicating their stability guarantees, whereas a lack of stability assurance raises the risk of critical damage or even complete failure under a loss of control. In this study, we improve a Robust, Optimal, Safe and Stability Guaranteed Training (ROSS-GT) method of [1] to design an NN controller for a quadcopter flight control. The approach ensures closed-loop system stability by finding a Lyapunov function, and providing a safe initial state domain that remains invariant under the control and guarantees stability to an equilibrium within it. Stability guaranteeing constraints are derived from the sector bound of the system nonlinearity and of its parameters and disturbance variations, in the form of a Lipschitz bound for a NN control. The control performance is further optimized by searching over the class of stability-guaranteeing controllers to minimize the reference tracking error and the control costs.

Figures

Figures reproduced from arXiv: 2412.14003 by the authors.

Figure 1
Figure 1. Controller with nominal vs. NN components [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 2
Figure 2. Nominal controller architecture The control inputs are set to be the motor thrusts: ui = ktω 2 i , ∀i ∈ {1, 2, 3, 4}, and these control inputs are allocated from the required thrust and torques using:     u1 u2 u3 u4     =     1 1 1 1 − √ l 2 − √ l 2 √ l 2 √ l 2 − √ l 2 √ l 2 √ l 2 − √ l 2 kd/kt −kd/kt kd/kt −kd/kt     −1    Td τϕ,d τθ,d τψ,d     . For Position Control: xe = xd − x, ye = yd −… view at source ↗
Figure 3
Figure 3. illustrates this actor-critic framework used in our training of optimal NN controller. The environment models the quadcopter dynamics along with the nominal controller described in Equation (10). The NN controller provides control inputs that are fed into the environment (nominally controlled quadcopter), which executes the control and returns a “reward” value depending on the current state, and also advances the st… view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Actor and Critic networks for quadcopter control [PITH_FULL_IMAGE:figures/full_fig_p008_4.png]
Figure 5
Figure 5. Figure 5: Reference position control input These inputs highlight the NN controller’s role in dynamically adjusting the system’s behavior to minimize position errors [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 6
Figure 6. Figure 6: Results of reference position tracking [PITH_FULL_IMAGE:figures/full_fig_p010_6.png]
Figure 7
Figure 7. Figure 7: Results of reference trajectory tracking [PITH_FULL_IMAGE:figures/full_fig_p011_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

26 extracted references · 12 canonical work pages

  1. [5]

    Efficient and accurate estimation of lipschitz constants for deep neural networks,

    M. Fazlyab, A. Robey, H. Hassani, M. Morari, and G. J. Pappas, “Efficient and accurate estimation of lipschitz constants for deep neural networks,” 2023. [Online]. Available: https://arxiv.org/abs/1906.04893 (a) Position with π = π0 (b) Position with π = π0 + πN N (c) Velocity with π = π0 (d) Velocity with π = π0 + πN N (e) Positional error with π = π0 (f...

    arXiv 2023

  2. [1]

    Robust Stability of Neural Network-controlled Nonlinear Systems with Parametric Variability

    S. Talukder and R. Kumar, “Robust stability of neural network- controlled nonlinear systems with parametric variability,” 2022. [Online]. Available: https://arxiv.org/abs/2109.05710

    work page Pith review arXiv 2022

  3. [2]

    Learn by Observation: Imitation Learning for Drone Patrolling from Videos of A Human Navigator

    Y . Fan, S. Chu, W. Zhang, R. Song, and Y . Li, “Learn by observation: Imitation learning for drone patrolling from videos of a human navigator,” 2020. [Online]. Available: https://arxiv.org/abs/2008.13193

    work page Pith review arXiv 2020

  4. [3]

    Visual imitation learning for robot manipulation,

    M. Sieb, “Visual imitation learning for robot manipulation,” Master’s thesis, Carnegie Mellon University, Pittsburgh, PA, May. 2019

    work page 2019

  5. [4]

    R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. The MIT Press, 2018. [Online]. Available: http://incompleteideas.net/book/the-book-2nd.html

    2018

  6. [6]

    Control of a quadrotor with reinforcement learning,

    J. Hwangbo, I. Sa, R. Siegwart, and M. Hutter, “Control of a quadrotor with reinforcement learning,” IEEE Robotics and Automation Letters , vol. 2, DOI 10.1109/lra.2017.2720851, no. 4, Oct. 2017. [Online]. Available: http://dx.doi.org/10.1109/LRA.2017.2720851

    arXiv 2017

  7. [7]

    Reinforcement learning for uav attitude control,

    W. Koch, R. Mancuso, R. West, and A. Bestavros, “Reinforcement learning for uav attitude control,” ACM Trans. Cyber-Phys. Syst. , vol. 3, DOI 10.1145/3301273, no. 2, 2019. [Online]. Available: https://doi.org/10.1145/3301273

    doi:10.1145/3301273 2019

  8. [8]

    Sim- to-real quadrotor landing via sequential deep q-networks and domain randomization,

    R. Polvara, M. Patacchiola, M. Hanheide, and G. Neumann, “Sim- to-real quadrotor landing via sequential deep q-networks and domain randomization,” Robotics, vol. 9, no. 1, p. 8, 2020

    work page 2020

Show all 26 references
  1. [9]

    Robust deep reinforcement learning for quadcopter control,

    A. M. Deshpande, A. A. Minai, and M. Kumar, “Robust deep reinforcement learning for quadcopter control,” 2021. [Online]. Available: https://arxiv.org/abs/2111.03915

    2021 arXiv

  2. [10]

    Lyapunov-stable neural-network control,

    H. Dai, B. Landry, L. Yang, M. Pavone, and R. Tedrake, “Lyapunov-stable neural-network control,” 2021. [Online]. Available: https://arxiv.org/abs/2109.14152

    2021 arXiv

  3. [11]

    Neural lyapunov control,

    Y .-C. Chang, N. Roohi, and S. Gao, “Neural lyapunov control,” 2022. [Online]. Available: https://arxiv.org/abs/2005.00611

    2022 arXiv

  4. [12]

    Neural lyapunov control of unknown nonlinear systems with stability guarantees,

    R. Zhou, T. Quartz, H. D. Sterck, and J. Liu, “Neural lyapunov control of unknown nonlinear systems with stability guarantees,” 2022. [Online]. Available: https://arxiv.org/abs/2206.01913

    2022 arXiv

  5. [13]

    Neural lyapunov control for discrete-time systems,

    J. Wu, A. Clark, Y . Kantaros, and Y . V orobeychik, “Neural lyapunov control for discrete-time systems,” 2023. [Online]. Available: https://arxiv.org/abs/2305.06547

    2023 arXiv

  6. [14]

    Lyapunov-stable neural control for state and output feedback: A novel formulation,

    L. Yang, H. Dai, Z. Shi, C.-J. Hsieh, R. Tedrake, and H. Zhang, “Lyapunov-stable neural control for state and output feedback: A novel formulation,” 2024. [Online]. Available: https://arxiv.org/abs/2404.07956

    2024 arXiv

  7. [15]

    Control-theoretic analysis of smoothness for stability-certified reinforcement learning,

    M. Jin and J. Lavaei, “Control-theoretic analysis of smoothness for stability-certified reinforcement learning,” in 2018 IEEE Conference on Decision and Control (CDC) , DOI 10.1109/CDC.2018.8618996, 2018

    2018

  8. [16]

    Safety verification and robustness analysis of neural networks via quadratic constraints and semidefinite programming,

    M. Fazlyab, M. Morari, and G. J. Pappas, “Safety verification and robustness analysis of neural networks via quadratic constraints and semidefinite programming,” IEEE Transactions on Automatic Control , vol. 67, DOI 10.1109/TAC.2020.3046193, no. 1, pp. 1–15, 2022

  9. [17]

    Stability analysis using quadratic con- straints for systems with neural network controllers,

    H. Yin, P. Seiler, and M. Arcak, “Stability analysis using quadratic con- straints for systems with neural network controllers,” IEEE Transactions on Automatic Control, vol. 67, DOI 10.1109/TAC.2021.3069388, no. 4, pp. 1980–1987, 2022

    2021

  10. [18]

    Model- ing and pid cascade control of a quadcopter for trajectory tracking,

    E. A. Paiva, J. C. Soto, J. A. Salinas, and W. Ipanaqu ˜A©, “Model- ing and pid cascade control of a quadcopter for trajectory tracking,” in 2015 CHILEAN Conference on Electrical, Electronics Engineer- ing, Information and Communication Technologies (CHILECON) , DOI 10.1109/Ch...

    2015

  11. [19]

    Khalil, Nonlinear Control , ser

    H. Khalil, Nonlinear Control , ser. Always Learning. Pear- son, 2014. [Online]. Available: https://books.google.com/books?id= -WbjoAEACAAJ

    2014

  12. [20]

    Regularisation of neural networks by enforcing lipschitz continuity,

    H. Gouk, E. Frank, B. Pfahringer, and M. J. Cree, “Regularisation of neural networks by enforcing lipschitz continuity,” 2020. [Online]. Available: https://arxiv.org/abs/1804.04368

    2020 arXiv

  13. [21]

    3DR Iris - the ready to fly UAV Quadcopter — arducopter.co.uk,

    “3DR Iris - the ready to fly UAV Quadcopter — arducopter.co.uk,” https://www.arducopter.co.uk/iris-quadcopter-uav.html, [Accessed 09- 12-2024]

    2024

  14. [22]

    Aerodynamic effects compensation on multi-rotor uavs based on a neural network control allocation approach,

    S. P. Madruga, A. H. B. M. Tavares, S. O. D. Luiz, T. P. do Nascimento, and A. M. N. Lima, “Aerodynamic effects compensation on multi-rotor uavs based on a neural network control allocation approach,” IEEE/CAA Journal of Automatica Sinica , vol. 9, DOI 10.1109/JAS.2021.1004266...

  15. [23]

    Proximal policy optimization algorithms,

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, and O. Klimov, “Proximal policy optimization algorithms,” 2017. [Online]. Available: https://arxiv.org/abs/1707.06347

    2017 arXiv

  16. [24]

    High-dimensional continuous control using generalized advantage estimation,

    J. Schulman, P. Moritz, S. Levine, M. Jordan, and P. Abbeel, “High-dimensional continuous control using generalized advantage estimation,” 2018. [Online]. Available: https://arxiv.org/abs/1506.02438

    2018 arXiv

  17. [25]

    Trust region policy optimization,

    J. Schulman, S. Levine, P. Moritz, M. I. Jordan, and P. Abbeel, “Trust region policy optimization,” 2017. [Online]. Available: https: //arxiv.org/abs/1502.05477

    2017 arXiv

  18. [26]

    Minimum snap trajectory generation and control for quadrotors,

    D. Mellinger and V . Kumar, “Minimum snap trajectory generation and control for quadrotors,” in 2011 IEEE International Conference on Robotics and Automation, DOI 10.1109/ICRA.2011.5980409, pp. 2520– 2525, 2011

    2011

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.