Pith. sign in

REVIEW 4 major objections 5 minor 95 references

Precision positioning in free-space optical communication systems via PID control tuned by RL

T0 review · 4 major / 5 minor · reviewed 2026-08-01 · deepseek-v4-flash

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

Pith's one-line read The paper shows that a DDPG reinforcement-learning agent, acting over a 200-km UDP link, can tune six PID coefficients of a physical optical deflector so that radial tracking-error range falls from 119 to 82 (31%) on a dynamic pseudo-random

desk verdict A real distributed RL-PID tuning experiment on a physical deflector, with an honest negative result — but the 31% improvement is a single best-of-5000 draw with no stats, so judge it as a feasibility demo. read the letter →

arxiv 2607.15910 v1 pith:IGIV2THL submitted 2026-07-17 physics.ins-det

K. Prikhodko , S. Kuznetsov , S. Vorobey , A. Katanskiy , V. Balakirev , A. Reutov This is my paper
classification physics.ins-det
keywords free-spaceopticalcommunicationPIDcontrolreinforcementlearningDDPGdeflectorbeamalignmentdistributedpositioningerror
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 tries to establish that reinforcement learning can be used as an experimental, data-driven tuning layer for conventional PID controllers in a real free-space optical positioning system, rather than as a replacement for the controller. A DDPG agent interacts with a physical optical deflector that moves the tip of an optical fiber in the focal plane; agent and device are 200 km apart and exchange data through UDP packets, so the learning process includes genuine communication delays and losses. After 5000 sessions the agent found two fixed six-coefficient sets. In a dynamic pseudo-random trajectory test, the better set reduced the radial positioning-error range from 119 to 82 (31%) and its standard deviation from 15 to 12; in a constant-target test, neither RL-tuned set improved on manual tuning. The claim is therefore not universal superiority, but feasibility and a measurable gain in one important operating regime.

What carries the argument

The load-bearing mechanism is a Deep Deterministic Policy Gradient (DDPG) agent operating in a closed loop over a distributed network. Its action is a six-dimensional vector of PID coefficients (Pp, Ip, Dp for the position loop, Pv, Iv, Dv for the velocity loop), transmitted through UDP to the optical deflector. Its state is a sequence of 362 measured and target coordinate pairs, its reward is the sum of radial distances between position and target over those 362 samples, and epsilon-greedy noise provides exploration. This arrangement lets the agent perform black-box, policy-guided search over controller gains directly on hardware, including the device's nonlinearities, sensor noise, and net

What would settle it

Run the selected coefficient set aopt,1 on a trajectory family that differs from the training PRS (for example, a chirp, a recorded platform-vibration sequence, or band-limited noise with a different cutoff), and compare radial error range against the baseline. If the range reduction disappears, reverses, or is not reproduced across repeated trials, the reported 31% gain is specific to the generator rather than a general property of the tuned gains.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is that DDPG-assisted PID tuning is feasible on a physical stand: the RL agent adjusted the six gains of cascaded position and velocity PID controllers over thousands of real communication sessions, and the selected coefficients aopt,1 improved dynamic tracking (radial range 82 vs 119, sigma 12 vs 15) compared with the manually tuned baseline. For the stationary zero-target regime, the RL-tuned sets did not improve radial error range. The authors interpret this as evidence for the potential of RL-based experimental PID tuning in dynamic tasks and as a warning that the training trajectory and reward must match the intended operating regime.

Load-bearing premise

The pseudo-random target trajectory used for both training and testing is generated by filtering uniform polar noise through a low-pass filter whose frequency response is matched to the optical deflector; if this generator does not represent real operating conditions, or if the evaluation uses the same signal family as training, the observed 31% gain is an artifact of optimizing and testing within the same narrow signal class.

Editorial extensions

If this is right

  • If the result holds, RL-based tuning can be done in the field from a remote location, enabling automated commissioning of precision positioning systems without a local engineer.
  • The 31% range reduction in dynamic tracking implies the selected PID gains better compensate the deflector's frequency response for fast target sequences, which could improve coupling stability in FSO links under vibration.
  • The failure on constant targets suggests single-regime rewards can overfit: any deployment should specify a mixture of operating modes in the reward to avoid degrading static accuracy.
  • The method is general: the same DDPG-over-UDP approach can be applied to other cascaded PID plants, provided a safe exploration range and a self-preserving penalty are defined.

Reading between the lines

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

  • The improvement is only shown for the same signal class used in training; the paper never tests a different trajectory family, so the 31% figure may partly reflect specialization to the PRS generator rather than a robust improvement.
  • A plausible next experiment is to train on a reward that mixes dynamic and stationary regimes, which the paper itself suggests; this would likely recover zero-target performance while retaining dynamic gains.
  • Because the agent only updates coefficients between sessions, the approach is episodic black-box optimization; interpreting DDPG as online is a stretch—this is closer to guided search over a low-dimensional parameter space, and simpler baselines such as Bayesian optimization could provide a stronger benchmark.
  • The remote 200-km setup implies that long-latency, lossy channels do not prevent data-driven tuning; this could extend to tuning controllers on platforms that are hard to access physically, like telescopes or orbital systems.
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. The paper describes a DDPG agent that tunes six coefficients of cascaded position and velocity PID controllers for an optical deflector in a free-space optical positioning stand. Training is performed over 5000 remote UDP communication sessions against the physical plant. Two coefficient sets are selected from the recorded training history as the highest-reward actions. These sets, along with a manually tuned baseline, are evaluated on two target regimes: a constant {0,0} target and a 12-kHz pseudo-random trajectory matched to the OD frequency response (Eq. 3). The best RL-tuned set yields a 31% reduction in radial error range (119 to 82) and a standard-deviation reduction (15 to 12) for the PRS, while no improvement over baseline is observed for the static target. The paper concludes that DDPG-assisted PID tuning is feasible for the tested dynamic regime but does not generalize universally.

Significance. The experimental setup is unusual and valuable: the RL agent and the physical plant are separated by 200 km with UDP transport, so the training includes real network delays and failures. The authors are honest about the null result for a static target and explicitly state that the coefficients are not optimal for all operating conditions. Strengths include a real-hardware test rather than simulation, a clear description of the stand and protocol, and an explicit call for repeatability experiments. However, the central quantitative claim (31% improvement) rests on a best-of-5000 selection and a single evaluation on the same trajectory family used in training. The missing statistical replication and the lack of a held-out trajectory class prevent the paper from establishing that the measured improvement is systematic rather than a selection artifact.

major comments (4)
  1. [Section III.B, Table II, Table IV] The two coefficient sets are chosen as the maximum-reward actions from 5000 training sessions, and each session uses a fresh PRS from Eq. (3) with noisy reward (UDP failures, sensor noise, random target generation). The reported D_R reduction (119→82) is from a single 65535-point PRS test. With no repeated independent evaluations, this is exactly the kind of best-of-N selection bias that would make a noisy sample look better than its true mean. Please report the number of repetitions behind the 'Variance of D_R' column, provide confidence intervals, and include a paired statistical comparison against the baseline.
  2. [Section III.B, Eq. (3), Table IV] The PRS used for training and the PRS used for testing are generated by the same stochastic process: a uniform polar random vector followed by a low-pass filter whose frequency response is 'identical to the frequency response of the OD.' Thus the selected gains are optimized for the same signal class on which they are evaluated. To support a general dynamic-regime claim, test on a distinct trajectory family (e.g., chirp, multi-sine, or recorded operational disturbance) or at least on independent realizations from a different distribution. The static-target test is a useful contrast but is only one alternative regime.
  3. [Section II, Eq. (2), Fig. 3 caption] Eq. (2) defines the reward as the sum of Euclidean distances, but the Fig. 3 caption states that a 'self-preserving penalty' is applied when the error exceeds the baseline PID error, and that a penalty is also applied for connection failures. This penalty is not in Eq. (2), so the reward actually used for action selection is not the quantity defined in the text. Since the coefficient sets are selected by maximum reward, the paper must specify the full reward function, including the penalty and its scale.
  4. [Section IV, Table IV] The units of σ and D are not stated (presumably micrometers or device-specific units). More importantly, 'Variance of D_R' implies replicate measurements, but the number of repetitions is never given. Without this information, the statement that a_opt,1 is 'more optimal' than baseline because its D_R variance is halved is not statistically testable. Report the number of repeated runs, the distribution of D_R, and an appropriate significance test.
minor comments (5)
  1. [Throughout] Grammar and typos: 'authors was investigated' (Section I), 'is chose' (Section II), 'position positions' (Section III.A). A careful proofread is needed.
  2. [Section II, Eq. (1)] The notation Q(a_t, s_t) is nonstandard; DDPG typically uses Q(s, a). Also, showing a max over a_t' ignores the target-network and deterministic-policy details; please clarify the notation.
  3. [Section III.B, Eq. (3)] The formula y_target(t_i) = U(t_i) + P(t_i) is presented with 'U(x) is a step periodic function' and 'P(x) is a pseudo-random sequence'. For the static target, the text says the target is {0,0}, which is not a special case of this formula; clarify how the PRS and the constant target relate to Eq. (3).
  4. [Figure 3] The vertical axis label '-101' appears to be a typo; also, the log-scale normalization should be defined explicitly.
  5. [References] Several references are incomplete (e.g., [10] lacks venue/pages, [13] is an arXiv preprint without arXiv number, [1] is a Russian-language proceedings paper). Please complete the bibliographic information.

Circularity Check

0 steps flagged · score 1.0 of 10

No circular derivation: the 31% D_R reduction is an empirical measurement distinct from the training reward, with an honest negative control; the only self-citation (Ref. [1]) is hardware-descriptive and non-load-bearing.

full rationale

This is an empirical demonstration rather than a formal derivation, so the reduction-based circularity patterns do not apply. The RL-tuned coefficient sets are selected as the maximum-reward actions from 5000 recorded sessions (Section III.B, Table II), where the episodic reward is the summed radial distance r = k_r Σ √((x_pos−x_tgt)²+(y_pos−y_tgt)²) (Eq. 2). The headline claim, however, is the range D_R and σ_R of the radial error measured on a fresh 65535-point PRS (Section IV, Table IV). D_R is a max−min statistic, a different functional of the error than the summed-distance reward used for selection, so the 31% reduction (119 to 82) is not forced by construction. This is corroborated internally: the second selected set improves D_R by only 13.4%, and neither set improves D_R for a constant zero target — a negative control the paper reports honestly. Training and test trajectories are both drawn from Eq. (3) (U + PRS through a low-pass filter 'identical to the frequency response of the OD'), which limits generalization to other regimes but is standard train/evaluate-on-the-same-distribution practice, not circularity; the paper itself disclaims universality ('they do not establish a universal improvement under all operating conditions') and calls for repeatability experiments. The only self-citation, Ref. [1] (co-author S. N. Kuznetsov), supports the hardware description ('up to 12 kHz') and is not load-bearing for the RL-tuning claim. The flagged weaknesses — best-of-5000 selection bias and an unreported number of repetitions behind the 'Variance of D_R' column — are statistical-robustness gaps, not reduction-to-inputs, and belong under correctness risk.

Assumptions & free parameters 8 free parameters · 5 assumptions · 0 invented entities

The quantitative claim rests on six RL-selected PID gains and the DDPG hyperparameter choices, all fitted or chosen by hand. The axioms are standard RL convergence assumptions plus domain assumptions about the OD frequency response, channel stationarity, and metric stability. No new physical entities are postulated; the self-preserving penalty mentioned in the Fig. 3 caption is a reward-shaping term, not an entity.

free parameters (8)
  • P_p (position proportional gain), a_opt,1 = 4.602
    Selected as one of the two reward-maximizing coefficient sets from 5000 training sessions (Table II). The central 31% result depends on this value.
  • I_p (position integral gain), a_opt,1 = 0.08649
    Reward-selected coefficient; part of the a_opt,1 set that produces the 31% improvement.
  • D_p (position derivative gain), a_opt,1 = 12.29
    Reward-selected coefficient; part of the a_opt,1 set.
  • P_v (velocity proportional gain), a_opt,1 = 3.238
    Reward-selected coefficient; part of the a_opt,1 set.
  • I_v (velocity integral gain), a_opt,1 = 0.0004861
    Reward-selected coefficient; part of the a_opt,1 set.
  • D_v (velocity derivative gain), a_opt,1 = 14.34
    Reward-selected coefficient; part of the a_opt,1 set.
  • Second reward-selected set a_opt,2 = Pp=4.676, Ip=0.08508, Dp=12.95, Pv=3.119, Iv=0.0004785, Dv=15.32
    Second candidate set; its much larger variance of D_R (198 vs 41) shows that the reported improvement is sensitive to which training-history point is chosen.
  • DDPG hyperparameters (network sizes 80/60, discount factor gamma, epsilon schedule, ADAM settings, reward normalization = not fully specified
    Hand-chosen or taken from Lillicrap et al. without exact values; these choices can materially affect whether tuning succeeds, and the paper does not report them.
assumptions (5)
  • standard math Bellman equation converges to an optimal policy for the described Q-learning/DDPG setup
    Section II Eq. (1); standard, but the paper applies it to a semi-Markov process with nonstationary communication without proof.
  • domain assumption The OD frequency response is constant and correctly described by the low-pass filter used to generate PRS targets
    Section III.B: 'the frequency response of which was identical to the frequency response of the OD'. This is central to the representativeness of both training and test trajectories.
  • domain assumption The physical stand and UDP channel are stationary enough across 5000 training sessions and the final tests
    Training and test happen at different times; no drift, thermal, or channel-condition data are provided.
  • domain assumption Radial error range D_R and standard deviation are stable performance metrics despite unreported number of test repetitions
    Table IV reports variance of D_R but no N, confidence intervals, or significance tests.
  • domain assumption DDPG with two hidden layers of sizes 80 and 60 and Lillicrap hyperparameters is suitable for PID tuning on this plant
    Algorithm choice is justified by literature citations, not by ablation or comparison on this specific stand.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Precision positioning in free-space optical communication systems via PID control tuned by RL." pith.science (2026). https://pith.science/paper/IGIV2THL

@misc{pith2026260715910,
  author       = {Pith},
  title        = {Pith review of: Precision positioning in free-space optical communication systems via PID control tuned by RL},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IGIV2THL}},
  note         = {Machine review of arXiv:2607.15910}
}
abstract

Accurate positioning of optical components is essential for maintaining beam alignment in free-space optical (FSO) communication systems. This work investigates reinforcement-learning-assisted tuning of cascaded position and velocity PID controllers for an optical deflector that moves the end of an optical fiber in the focal plane of an optical system. A Deep Deterministic Policy Gradient (DDPG) agent adjusts six PID coefficients through interaction with a physical experimental stand. The stand supports target-coordinate updates of up to $12$ kHz, while the agent and the controlled device are located approximately $200$ km apart and exchange data over UDP. After $5000$ training sessions, two fixed coefficient sets are selected and compared with a manually tuned baseline. For a pseudo-random target trajectory, the best RL-tuned set reduces the range of the radial positioning error from $119$ to $82$, corresponding to a $31\%$ reduction, and decreases its standard deviation from $15$ to $12$. For a constant zero target, the RL-tuned sets do not improve the radial error range. The results demonstrate the potential of DDPG for experimental PID tuning in dynamic positioning tasks and indicate the need for multi-regime optimization to achieve consistent performance under different operating conditions.

Figures

Figures reproduced from arXiv: 2607.15910 by the authors.

Figure 2
Figure 2. FIG. 2. Scheme of an optical deflector. 1 is a permanent magnet, 2 [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 1
Figure 1. FIG. 1. Scheme of the experimental stand for optimization of PID [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗
Figure 3
Figure 3. FIG. 3. Reward [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

95 extracted references · 2 canonical work pages

  1. [1]

    N. D. Birell and P. C. W. Davies , year = 1982, title =

    1982

  2. [2]

    R. P. Feynman. Phys.\ Rev. 1954

    1954

  3. [3]

    Einstein and Yu Podolsky and N

    A. Einstein and Yu Podolsky and N. Rosen. Phys.\ Rev. 1935

    1935

  4. [4]

    Berman, Jr., G. P. and Izrailev, Jr., F. M. Stability of nonlinear modes. Physica D. 1983

    1983

  5. [5]

    E. B. Davies and L. Parns. Trapped modes in acoustic waveguides. Q. J. Mech. Appl. Math. 1988

    1988

  6. [6]

    Edward Witten. 2001. hep-th/0106109

    arXiv 2001

  7. [7]

    E. Beutler. Williams Hematology. 1994

    1994

  8. [8]

    Donald E. Knuth. Fundamental Algorithms. 1973b 1973

    1973

Show all 95 references
  1. [9]

    J. S. Smith and G. W. Johnson. Philos. Trans. R. Soc. London, Ser. B. 2005

    2005

  2. [10]

    W. J. Smith and T. J. Johnson and B. G. Miller. Surface chemistry and preferential crystal orientation on a silicon surface

  3. [11]

    V. K. Smith and K. Johnson and M. O. Klein. Surface chemistry and preferential crystal orientation on a silicon surface

  4. [12]

    Lower Bounds for Wishful Research Results

    Ulrich \" U nderwood and Ned \ N et and Paul \= P ot. Lower Bounds for Wishful Research Results

  5. [13]

    M. P. Johnson and K. L. Miller and K. Smith. 2007

    2007

  6. [14]

    AIP Conf. Proc. 2007

    2007

  7. [15]

    Fifteenth Annual

    Proc. Fifteenth Annual

  8. [16]

    Y. Burstyn. Proceedings of the 5th International Molecular Beam Epitaxy Conference, Santa Fe, NM. 2004

    2004

  9. [17]

    Proceedings of the 2003 Particle Accelerator Conference, Portland, OR, 12-16 May 2005. 2001

    2003

  10. [18]

    A. G. Agarwal. Proceedings of the Fifth Low Temperature Conference, Madison, WI, 1999. Semiconductors. 2001

    1999

  11. [19]

    R. Smith. Hummingbirds are our friends

  12. [20]

    J. Smith. Proc. SPIE. 2007

    2007

  13. [21]

    An O(n n / \! n) Sorting Algorithm

    Tom T \' e rrific. An O(n n / \! n) Sorting Algorithm

  14. [22]

    Mastering Thesis Writing

    \' E douard Masterly. Mastering Thesis Writing

  15. [23]

    S. R. Kawa and S.-J. Lin. J. Geophys. Res. 2003

    2003

  16. [24]

    Phidias Phony-Baloney

    F. Phidias Phony-Baloney. Fighting Fire with Fire: Festooning F rench Phrases

  17. [25]

    Donald E. Knuth. Seminumerical Algorithms. 1973c 1981

    1981

  18. [26]

    Jill C. Knvth. The Programming of Computer Art

  19. [28]

    Opechowski and R

    W. Opechowski and R. Guccione. Introduction to the Theory of Normal Metals. Magnetism

  20. [29]

    Opechowski and R

    W. Opechowski and R. Guccione. Introduction to the Theory of Normal Metals. Magnetism. 1965

    1965

  21. [30]

    J. M. Smith. Molecular Dynamics. 1980

    1980

  22. [31]

    V. E. Zakharov and A. B. Shabat. Exact theory of two-dimensional self-focusing and one-dimensional self-modulation of waves in nonlinear media. Zh. Eksp. Teor. Fiz. 1971

    1971

  23. [32]

    Ballagh and C.M

    R. Ballagh and C.M. Savage. Bose-Einstein condensation: from atomic physics to quantum fluids. Proceedings of the 13th Physics Summer School. 2000. cond-mat/0008070

    2000 arXiv

  24. [33]

    Daniel D. Lincoll. Semigroups of Recurrences. High Speed Computer and Algorithm Organization

  25. [34]

    Oaho and Jeffrey D

    Alfred V. Oaho and Jeffrey D. Ullman and Mihalis Yannakakis. On Notions of Information Transfer in VLSI Circuits. Proc. Fifteenth Annual ACM

  26. [35]

    The Definitive Computer Manual

    Larry Manmaker. The Definitive Computer Manual

  27. [36]

    author author A. A. \ Boev , author M. Y. \ Kernosov , author S. N. \ Kuznetsov , author A. A. \ Parshin , author S. Y. \ Polyakov , \ and\ author S. E. \ Shirobakin ,\ in\ @noop booktitle Trudy XXIX Mezhdunarodnoj Konferencii «Lazerno-informacionnye tekhnologii» [Proceedings ...

    2021

  28. [37]

    Borase , author D

    author author R. Borase , author D. Maghade , author S. Sondkar , \ and\ author S. Pawar ,\ 10.1007/s40435-020-00665-4 journal journal International Journal of Dynamics and Control \ volume 9 ( year 2021 ),\ 10.1007/s40435-020-00665-4 NoStop

    2021 doi

  29. [38]

    author author M. A. \ Johnson \ and\ author M. H. \ Moradi ,\ @noop title PID control \ ( publisher Springer ,\ year 2005 ) NoStop

    2005

  30. [39]

    author author J. G. \ Ziegler \ and\ author N. B. \ Nichols ,\ @noop journal journal Transactions of the American society of mechanical engineers \ volume 64 ,\ pages 759 ( year 1942 ) NoStop

    1942

  31. [40]

    author author A. E. \ Sallab et al. ,\ @noop journal journal Electronic Imaging \ volume 29 ,\ pages 70–76 ( year 2017 ) NoStop

    2017

  32. [41]

    author author B. R. \ Kiran et al. ,\ @noop title Deep reinforcement learning for autonomous driving: A survey , \ ( year 2021 ),\ http://arxiv.org/abs/2002.00444 arXiv:2002.00444 [cs.LG] NoStop

    2021 arXiv

  33. [42]

    Wei , author Y

    author author T. Wei , author Y. Wang , \ and\ author Q. Zhu ,\ in\ @noop booktitle Proceedings of the 54th Annual Design Automation Conference 2017 ,\ series and number DAC '17 \ ( publisher Association for Computing Machinery ,\ address New York, NY, USA ,\ year 2017 ) NoStop

    2017

  34. [43]

    Mocanu et al

    author author E. Mocanu et al. ,\ @noop journal journal IEEE Transactions on Smart Grid \ volume 10 ,\ pages 3698 ( year 2019 ) NoStop

    2019

  35. [44]

    Bøhn , author E

    author author E. Bøhn , author E. M. \ Coates , author S. Moe , \ and\ author T. A. \ Johansen ,\ 10.1109/ICUAS.2019.8798254 title Deep reinforcement learning attitude control of fixed-wing uavs using proximal policy optimization , \ ( year 2019 ) NoStop

    2019

  36. [45]

    Qin , author W

    author author Y. Qin , author W. Zhang , author J. Shi , \ and\ author L. Jinglong \ ( year 2018 )\ pp.\ pages 1--6 NoStop

    2018

  37. [46]

    Ming , author X

    author author R. Ming , author X. Liu , author Y. Li , author Y. Yin , \ and\ author W. Zhang ,\ 10.1007/s10489-023-04876-y journal journal Applied Intelligence \ ( year 2023 ),\ 10.1007/s10489-023-04876-y NoStop

    2023 doi

  38. [47]

    Schulman , author F

    author author J. Schulman , author F. Wolski , author P. Dhariwal , author A. Radford , \ and\ author O. Klimov ,\ @noop title Proximal policy optimization algorithms , \ ( year 2017 a ),\ http://arxiv.org/abs/1707.06347 arXiv:1707.06347 [cs.LG] NoStop

    2017 arXiv

  39. [49]

    Carlucho , author M

    author author I. Carlucho , author M. De Paula , \ and\ author G. G. \ Acosta ,\ @noop journal journal ISA Transactions \ volume 102 ,\ pages 280 ( year 2020 ) NoStop

    2020

  40. [50]

    ,\ @noop journal journal 31st International Conference on Machine Learning, ICML 2014 \ volume 1 ( year 2014 ) NoStop

    author author Silver et al. ,\ @noop journal journal 31st International Conference on Machine Learning, ICML 2014 \ volume 1 ( year 2014 ) NoStop

    2014

  41. [51]

    author author T. P. \ Lillicrap et al. ,\ @noop title Continuous control with deep reinforcement learning , \ ( year 2019 ),\ http://arxiv.org/abs/1509.02971 arXiv:1509.02971 [cs.LG] NoStop

    2019 arXiv

  42. [52]

    author author R. S. \ Sutton , author D. Precup , \ and\ author S. Singh ,\ @noop journal journal Artificial intelligence \ volume 112 ,\ pages 181 ( year 1999 ) NoStop

    1999

  43. [53]

    Schulman et al

    author author J. Schulman et al. ,\ @noop title Trust region policy optimization , \ ( year 2017 b ),\ http://arxiv.org/abs/1502.05477 arXiv:1502.05477 [cs.LG] NoStop

    2017 arXiv

  44. [54]

    Haarnoja , author A

    author author T. Haarnoja , author A. Zhou , author P. Abbeel , \ and\ author S. Levine ,\ in\ https://proceedings.mlr.press/v80/haarnoja18b.html booktitle Proceedings of the 35th International Conference on Machine Learning ,\ series Proceedings of Machine Learning Research ,...

    2018

  45. [55]

    Nian , author J

    author author R. Nian , author J. Liu , \ and\ author B. Huang ,\ https://doi.org/10.1016/j.compchemeng.2020.106886 journal journal Computers & Chemical Engineering \ volume 139 ,\ pages 106886 ( year 2020 ) NoStop

    2020

  46. [56]

    Henderson , author R

    author author P. Henderson , author R. Islam , author P. Bachman , author J. Pineau , author D. Precup , \ and\ author D. Meger \ ( publisher AAAI Press ,\ year 2018 ) NoStop

    2018

  47. [57]

    Mnih et al

    author author V. Mnih et al. ,\ @noop journal journal Nature \ volume 518 ,\ pages 529 ( year 2015 ) NoStop

    2015

  48. [58]

    author author D. P. \ Kingma \ and\ author J. Ba ,\ @noop title Adam: A method for stochastic optimization , \ ( year 2017 ) NoStop

    2017

  49. [59]

    Deterministic Policy Gradient Algorithms , volume =

    Silver and others , year =. Deterministic Policy Gradient Algorithms , volume =

  50. [60]

    2019 , eprint=

    Continuous control with deep reinforcement learning , author=. 2019 , eprint=

    2019

  51. [61]

    and Moe, Signe and Johansen, Tor Ame , booktitle=

    Bøhn, Eivind and Coates, Erlend M. and Moe, Signe and Johansen, Tor Ame , booktitle=. Deep Reinforcement Learning Attitude Control of Fixed-Wing UAVs Using Proximal Policy optimization , year=

  52. [62]

    2017 , eprint=

    Trust Region Policy Optimization , author=. 2017 , eprint=

    2017

  53. [63]

    2017 , eprint=

    Proximal Policy Optimization Algorithms , author=. 2017 , eprint=

    2017

  54. [64]

    doi:10.48550/ARXIV.2112.15187 , author =

    Stability-Preserving Automatic Tuning of PID Control with Reinforcement Learning , publisher =. doi:10.48550/ARXIV.2112.15187 , author =

    doi arXiv

  55. [65]

    A review of PID control, tuning methods and applications , volume =

    Borase, Rakesh and Maghade, Dilip and Sondkar, Shilpa and Pawar, Sushant , year =. A review of PID control, tuning methods and applications , volume =. International Journal of Dynamics and Control , doi =

  56. [66]

    2005 , publisher=

    PID control , author=. 2005 , publisher=

    2005

  57. [67]

    An adaptive deep reinforcement learning approach for MIMO PID control of mobile robots , journal =

    Ignacio Carlucho and Mariano. An adaptive deep reinforcement learning approach for MIMO PID control of mobile robots , journal =. 2020 , issn =

    2020

  58. [68]

    Physical review , volume=

    On the theory of the Brownian motion , author=. Physical review , volume=. 1930 , publisher=

    1930

  59. [69]

    2018 , isbn =

    Deep Reinforcement Learning That Matters , author =. 2018 , isbn =

    2018

  60. [70]

    Computers & Chemical Engineering , volume =

    A review On reinforcement learning: Introduction and applications in industrial process control , author =. Computers & Chemical Engineering , volume =. 2020 , issn =

    2020

  61. [71]

    Proceedings of the 35th International Conference on Machine Learning , pages =

    Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor , author =. Proceedings of the 35th International Conference on Machine Learning , pages =. 2018 , editor =

    2018

  62. [72]

    Artificial intelligence , volume=

    Between MDPs and semi-MDPs: A framework for temporal abstraction in reinforcement learning , author=. Artificial intelligence , volume=. 1999 , publisher=

    1999

  63. [73]

    Qin, Yunxiao and Zhang, Weiguo and Shi, Jingping and Jinglong, Liu , year =

  64. [74]

    Morphing aircraft acceleration and deceleration task morphing strategy using a reinforcement learning method , journal =

    Ming, Ruichen and Liu, Xiaoxiong and Li, Yu and Yin, Yi and Zhang, Weiguo , year =. Morphing aircraft acceleration and deceleration task morphing strategy using a reinforcement learning method , journal =

  65. [75]

    Nature , keywords =

    Mnih, Volodymyr and others , description =. Nature , keywords =

  66. [76]

    Transactions of the American society of mechanical engineers , volume=

    Optimum settings for automatic controllers , author=. Transactions of the American society of mechanical engineers , volume=. 1942 , publisher=

    1942

  67. [77]

    2017 , eprint=

    A Deep Reinforced Model for Abstractive Summarization , author=. 2017 , eprint=

    2017

  68. [78]

    and Faria, Diego , year =

    Bird, Jordan and Ekárt, A. and Faria, Diego , year =. On the effects of pseudorandom and quantum-random number generators in soft computing , volume =

  69. [79]

    and Barto, Andrew G

    Sutton, Richard S. and Barto, Andrew G. , edition =. Reinforcement Learning: An Introduction , year =

  70. [80]

    2017 , archivePrefix=

    Adam: A Method for Stochastic Optimization , author=. 2017 , archivePrefix=

    2017

  71. [81]

    Quantum random number generators , author =. Rev. Mod. Phys. , volume =. 2017 , month =

    2017

  72. [82]

    Source-device-independent heterodyne-based quantum random number generator at 17 Gbps , volume=

    Avesani, Marco and others , year=. Source-device-independent heterodyne-based quantum random number generator at 17 Gbps , volume=. Nature Communications , publisher=

  73. [83]

    Maximization of Extractable Randomness in a Quantum Random-Number Generator , author =. Phys. Rev. Applied , volume =. 2015 , month =

    2015

  74. [84]

    Journal of Modern Optics , volume =

    André Stefanov and others , title =. Journal of Modern Optics , volume =. 2000 , publisher =

    2000

  75. [85]

    WITHDRAWN: Quantum cryptography: Public key distribution and coin tossing , volume =

    Bennett, Charles and Brassard, Gilles , year =. WITHDRAWN: Quantum cryptography: Public key distribution and coin tossing , volume =

  76. [86]

    and others , year=

    Lucamarini, M. and others , year=. Overcoming the rate–distance limit of quantum key distribution without quantum repeaters , volume=. Nature , publisher=

  77. [87]

    2019 , journal =

    Quantum Supremacy using a Programmable Superconducting Processor , author =. 2019 , journal =

    2019

  78. [88]

    2021 , eprint=

    Deep Reinforcement Learning for Autonomous Driving: A Survey , author=. 2021 , eprint=

    2021

  79. [89]

    Deep Reinforcement Learning framework for Autonomous Driving , volume=

    Sallab, Ahmad EL and others , year=. Deep Reinforcement Learning framework for Autonomous Driving , volume=. Electronic Imaging , publisher=

  80. [90]

    Nature , keywords =

    Silver, David and and others , biburl =. Nature , keywords =

  81. [91]

    On-Line Building Energy Optimization Using Deep Reinforcement Learning , year=

    Mocanu, Elena and others , journal=. On-Line Building Energy Optimization Using Deep Reinforcement Learning , year=

  82. [92]

    Proceedings of the 54th Annual Design Automation Conference 2017 , articleno =

    Wei, Tianshu and Wang, Yanzhi and Zhu, Qi , title =. Proceedings of the 54th Annual Design Automation Conference 2017 , articleno =. 2017 , isbn =

    2017

  83. [93]

    2021 , booktitle =

    Boev, A A and Kernosov, M Y and Kuznetsov, S N and Parshin, A A and Polyakov, S Y and Shirobakin, S E , title =. 2021 , booktitle =

    2021

  84. [94]

    2017 , eprint=

    Seq2SQL: Generating Structured Queries from Natural Language using Reinforcement Learning , author=. 2017 , eprint=

    2017

  85. [95]

    Quantum machine learning , volume=

    Biamonte, Jacob and others , year=. Quantum machine learning , volume=. Nature , publisher=

  86. [96]

    Quantum machine learning: A classical perspective , volume =

    Ciliberto, Carlo and and others , year =. Quantum machine learning: A classical perspective , volume =

  87. [97]

    and others , title =

    Bassham, Lawrence E. and others , title =. 2010 , publisher =

    2010

Pith tools

Reviewed August 1, 2026 · model on record in the stance chip above.