Pith. sign in

REVIEW 3 major objections 6 minor 34 references

Grower-in-the-Loop Interactive Reinforcement Learning for Greenhouse Climate Control

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Interactive RL with imperfect grower input can improve greenhouse climate control when the input steers action selection, but fails when it reshapes the reward.

desk verdict Solid simulation study whose qualitative comparison is plausible but whose headline numbers are undermined by beta tuning on the test set and missing uncertainty reporting. read the letter →

arxiv 2505.23355 v3 pith:SKTTS3QU submitted 2025-05-29 cs.LG math.OC

Maxiu Xiao , Jianglin Lan , Jingxin Yu , Weihong Ma , Qiuju Xie , Congcong Sun This is my paper
classification cs.LGmath.OC
keywords interactivereinforcementlearninggrower-in-the-loopgreenhouseclimatecontrolpolicyshapingrewardsharingimperfecthumaninputPPO
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

The paper asks whether interactive reinforcement learning, in which a grower's advice is mixed into training, can improve greenhouse climate control when that advice is imperfect. It tests three integration mechanisms (policy shaping, control sharing, and reward shaping) with simulated grower inputs in a lettuce greenhouse model. The central finding is that mechanisms influencing action selection improve profit by 8.4% and 6.8% over baseline PPO, while reward shaping decreases profit by 9.4%. The authors argue that this ordering follows from PPO's advantage estimation, because biased input that reshapes the reward corrupts the value baseline, whereas input that steers action selection injects exploration without destroying the reward signal. A sympathetic reader would care because grower domain knowledge is otherwise discarded, and this is a first demonstration that it can be reused even when imperfect.

What carries the argument

The central object is the mechanism by which grower input enters the PPO agent: reward shaping modifies the reward to r' = r + beta*F(s,a); control sharing overrides the agent's action selection with probability beta toward the action that maximizes estimated feedback; and policy shaping mixes the agent's action distribution with the grower-advice distribution using weight beta. Auxiliary neural networks estimate the grower's feedback or advice, and a discrepancy-based selective strategy, inspired by SafeDAgger, chooses which time steps receive input. The generalized advantage estimator (GAE) is the explanatory mechanism that makes action-selection methods robust to imperfect input, because underestimating the value function under biased guidance can accelerate policy improvement instead of distorting the learning objective.

What would settle it

Run the same three interactive RL mechanisms in the same simulated lettuce greenhouse but with advice collected from real, experienced growers, then measure profit over full 56-day test episodes; the central claim fails if reward shaping matches or exceeds policy shaping, or if policy shaping no longer improves profit over baseline PPO.

Watch

Extended reading notes

Core claim

The paper's central claim is that interactive RL with imperfect grower inputs can improve the performance of an RL agent in greenhouse climate control, provided the integration mechanism influences action selection rather than the reward function. Specifically, policy shaping with precise action advice achieves an 8.4% profit improvement over baseline PPO, control sharing achieves 6.8%, policy shaping with constraint advice achieves 3.1%, and reward shaping loses 9.4%. The mechanism difference is traced to the generalized advantage estimator: action-selection methods keep the reward signal intact and use the input to guide exploration, while reward shaping changes the objective the critic estimates, so biased input propagates into the value function and misguides learning.

Load-bearing premise

The load-bearing premise is that the simulated grower inputs, generated from the same optimal-control model that defines the training environment and deliberately biased toward short-term growth, are representative of how real growers would actually advise an RL agent.

Editorial extensions

If this is right

  • Greenhouse RL systems can leverage grower knowledge without requiring perfect advice, making human-in-the-loop automation more practical.
  • Practitioners should prefer policy shaping or control sharing over reward shaping when human inputs are noisy or biased.
  • The input weight beta must be tuned and annealed to zero during training, since too high an early weight slows later learning.
  • Action advice tolerates limited input availability better than feedback, especially with discrepancy-based selection, enabling query-efficient grower involvement.
  • Reward shaping should be avoided in PPO-based interactive RL unless the human input is nearly optimal, because imperfect rewards propagate into the value estimate.

Reading between the lines

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

  • If GAE is the causal mechanism, the ordering of the three methods may not transfer to value-based RL algorithms or to policy-gradient methods without advantage estimation, a scope the paper explicitly leaves open.
  • Real grower advice is likely more biased toward yield quality, risk aversion, and long-term crop health than the simulated short-term-growth signals; the qualitative advantage of action-selection mechanisms would likely persist, but the reported profit magnitudes may shrink or shift.
  • The discrepancy-based selective querying strategy could be combined with inverse reinforcement learning to ask growers only for advice on states where the agent's uncertainty is highest, improving sample efficiency further.
  • The 14-day truncated training episodes may favor short-sighted advice; training on full 56-day cycles could change the balance between methods that accelerate short-term reward and those that preserve long-term profit.
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

3 major / 6 minor

Summary. This paper proposes and evaluates three interactive RL mechanisms for greenhouse climate control: reward shaping, policy shaping, and control sharing, all built on a PPO baseline. The authors define three simulated grower input types (feedback, precise action advice, and constraint advice), analyze trade-offs among input availability, cognitive bias, latency, and knowledge level, and introduce auxiliary neural networks with a pool-based selection strategy to handle limited input availability. In a simulated Van Henten lettuce greenhouse, they report that policy shaping with precise advice improves test reward by 8.4%, control sharing by 6.8%, and policy shaping with constraints by 3.1%, while reward shaping decreases reward by 9.4% relative to baseline PPO. They conclude that action-selection mechanisms are robust to imperfect grower input while reward shaping is not, and they attribute this to PPO's GAE estimator. The paper also studies sensitivity to the input weight beta and to input availability.

Significance. The paper addresses a relevant and under-explored application: interactive RL for greenhouse climate control. It provides a useful taxonomy of input types and imperfect-input characteristics, and it implements three representative integration mechanisms in a standard PPO framework with 15 training seeds and seven winter test trajectories. If the reported ordering is robust, the practical recommendation to prefer action-selection mechanisms over reward shaping under imperfect human guidance would be valuable for greenhouse automation. The strengths are the careful decomposition of input characteristics, the explicit comparison of three incorporation mechanisms, and the analysis of beta sensitivity. The main uncertainty is not the derivation but the empirical validity of the headline comparison, because beta is selected on the test set and no interval estimates are provided; the representativeness of the simulated grower is also untested. These concerns are fixable and do not invalidate the framework proposed.

major comments (3)
  1. [Section 5.3, Table 9, Figure 6] The reported ranking is based on beta values 'chosen to be the ones that work best in the test' (Section 5.3). Because Table 9 reports each algorithm at its own best beta (0.2, 0.1, 0.05, 0.2) and Figure 6 shows that test reward is highly sensitive to beta (e.g., beta=0.5 degrades performance while beta=0.05-0.2 improves it), the headline improvements of +8.4%, +6.8%, +3.1% and the decline of -9.4% are best-case values computed on the same seven test trajectories used for evaluation. This selection procedure can inflate the gains and confounds the algorithm comparison with beta tuning; the claimed robustness ordering is therefore not established. Please either select beta on a validation split independent of the test trajectories, or report the full beta sweep for every algorithm and show that the ordering is stable across reasonable beta values.
  2. [Table 9 and Figure 4] Point estimates are reported without any measure of uncertainty. The paper trains each configuration with 15 seeds (Section 5.2) and evaluates on seven test trajectories (Section 5.1), but no standard deviations, confidence intervals, or significance tests are reported for the cumulative rewards or for the relative changes. Given the small test set and the demonstrated sensitivity to beta, the observed differences (e.g., 1.91 versus 2.07 Hfl/m2) may be within seed or trajectory noise. The authors should report per-seed and per-trajectory statistics, or at least standard errors, before claiming that the interactive methods improve over baseline and that reward shaping is worse.
  3. [Section 3.3 and the Appendix] The simulated grower inputs are generated from pre-calculated optimal values and ranges derived from the same Van Henten greenhouse model that defines the training and evaluation environment, with a deliberate bias toward maximizing current crop growth. The 'imperfect grower' is therefore a model-derived oracle with an injected bias structure, not an independent source of expertise or a sample of real grower behavior. The paper's conclusions about robustness to imperfect inputs and the ranking of the three mechanisms may depend on this specific bias pattern. This limitation should be stated prominently, and ideally the analysis should include a sensitivity study over alternative input error distributions (e.g., stochastic feedback, mis-specified ranges, or risk-averse advice) or a small real-grower validation. The current Section 6 limitation paragraph mentions only the choice of PPO, not this representativeness issue.
minor comments (6)
  1. [Equation (5)] Equation (5) is hard to parse: pi_agent(s,a) is a probability, so the phrase a = pi_agent(s,a) should be a = arg max over actions of pi_agent(s, .), and the second line should be normalized explicitly. Please rewrite for readability.
  2. [Throughout] The symbol beta is called the 'input weight (likelihood)' throughout. Since 'likelihood' has a standard statistical meaning, consider renaming it 'blending weight' or 'input weight' to avoid ambiguity.
  3. [Table 9] Please state whether the averages are over seeds, test trajectories, or both, and add the corresponding standard deviation in parentheses.
  4. [Section 5.5 and Figures 8-9] The reported 'test loss' is defined only in the text. Please give the exact loss definition (MSE for feedback, KL for advice) and state whether the curves are averaged over seeds and/or test trajectories.
  5. [Appendix] The model equations in the Appendix contain formatting ambiguities (for example, the term x1(t)^2 x3(t)/10^-5/2 is unclear). A cleaned equation set would improve reproducibility.
  6. [References] Reference [5] appears incomplete, and the DOI in reference [32] is line-broken; please check the bibliography.

Circularity Check

0 steps flagged · score 0.0 of 10

No derivation-style circularity: the reported algorithm comparison is an empirical simulation result, not an equation-level reduction to its inputs.

full rationale

The paper's central claims are empirical results from a simulation study, not a derivation chain. The simulated grower inputs (Section 3.3) are generated from pre-calculated optimal values of the same Van Henten greenhouse model that defines the training environment, with deliberate bias toward short-term growth; this is a modeling assumption that limits external validity, but it does not make any measured quantity equal to its own input by construction. Policy shaping and control sharing improve performance while reward shaping does not; that ordering is not forced by the definitions in Equations (4)-(6), since reward shaping could in principle have improved if the feedback had been helpful. The disclosed procedure of choosing the input weight beta as the value that 'work[s] best in the test' (Section 5.3) is a potential source of optimistic selection bias in the reported percentages, but it is a statistical internal-validity issue rather than circularity: Table 9 reports measured test rewards, and no fitted parameter is renamed as a prediction. The only self-citation, reference [16] for economic prices, is a data source and is not load-bearing for any claimed derivation. No equation is defined in terms of the quantity it purports to establish, and no uniqueness claim is imported from prior author work. Therefore the circularity score is 0.

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

The central claim depends on simulation fidelity and on the representativeness of the synthetic grower. Hyperparameters are tuned per algorithm, with the input weight beta selected on the test set. No code or dataset was released, so these assumptions cannot be independently checked.

free parameters (5)
  • Input weight beta (per algorithm) = 0.2 (policy shaping precise), 0.1 (policy shaping constraint), 0.05 (control sharing), 0.2 (reward shaping)
    Selected post hoc as the value with best test performance per algorithm in Table 9, a test-set fit.
  • Episode length (training) = 14 days (336 steps)
    Selected by tuning (Section 5.1) to balance rollout staleness and long-term optimization.
  • Entropy coefficient = 1e-2 (no inputs, reward shaping, policy shaping constraint), 1e-3 (policy shaping precise, control sharing)
    Adjusted per interactive algorithm to 'promote exploration', making cross-algorithm comparison uncontrolled.
  • Auxiliary network learning rate = 1e-4 (policy shaping precise), 1e-3 (reward shaping, control sharing)
    Set differently per algorithm in Table 8; a tuning choice.
  • Input count n per update interval = 2048 (full availability) down to 128
    Varied to test availability; full availability case in Table 9 uses 2048.
assumptions (5)
  • domain assumption The Van Henten 1994 lettuce greenhouse model is an accurate representation of real greenhouse climate and crop dynamics for the purpose of evaluating climate controllers.
    The environment, reward, and all training/test data are generated by this model (Appendix, Section 2.1). Any conclusions are conditional on this model being representative.
  • domain assumption The simulated grower inputs (feedback, precise advice, constraint advice) capture the characteristics of real grower inputs, including cognitive bias, limited availability, latency, and knowledge level.
    Section 3.3 constructs inputs from pre-calculated model-derived optimal values and ranges. There is no validation against human growers, so representativeness is assumed.
  • domain assumption PPO's generalized advantage estimation (GAE) is the mechanism that makes action-selection methods robust to imperfect inputs, and this conclusion generalizes across RL algorithms.
    The explanation in Section 5.3 attributes robustness to GAE, but only PPO is tested. The authors explicitly flag this as a limitation in Section 6.
  • domain assumption Winter weather data from Den Haag (training 2011-2020, testing 2021-2024) and the assumption of constant outdoor CO2 at 400 ppm are sufficient for evaluating greenhouse climate control.
    Section 5.1. The evaluation may not generalize to other climates or CO2 regimes.
  • domain assumption The economic reward function (lettuce price, heating cost, CO2 cost, constraint penalty) reflects real greenhouse economics.
    Equation (1) and Table 2; prices are taken from reference [16]. Absolute profit values are only meaningful if these prices are realistic.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Grower-in-the-Loop Interactive Reinforcement Learning for Greenhouse Climate Control." pith.science (2026). https://pith.science/paper/SKTTS3QU

@misc{pith2026250523355,
  author       = {Pith},
  title        = {Pith review of: Grower-in-the-Loop Interactive Reinforcement Learning for Greenhouse Climate Control},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SKTTS3QU}},
  note         = {Machine review of arXiv:2505.23355}
}
read the original abstract

Climate control is crucial for greenhouse production as it directly affects crop growth and resource use. Reinforcement learning (RL) has received increasing attention in this field, but still faces challenges, including limited training efficiency and high reliance on initial learning conditions. Interactive RL, which combines human (grower) input with the RL agent's learning, offers a potential solution to overcome these challenges. However, interactive RL has not yet been applied to greenhouse climate control and may face challenges related to imperfect inputs. Therefore, this paper aims to explore the possibility and performance of applying interactive RL with imperfect inputs into greenhouse climate control, by: (1) developing three representative interactive RL algorithms tailored for greenhouse climate control (reward shaping, policy shaping and control sharing); (2) analyzing how input characteristics are often contradicting, and how the trade-offs between them make grower's inputs difficult to perfect; (3) proposing a neural network-based approach to enhance the robustness of interactive RL agents under limited input availability; (4) conducting a comprehensive evaluation of the three interactive RL algorithms with imperfect inputs in a simulated greenhouse environment. The demonstration shows that interactive RL incorporating imperfect grower inputs has the potential to improve the performance of the RL agent. RL algorithms that influence action selection, such as policy shaping and control sharing, perform better when dealing with imperfect inputs, achieving 8.4% and 6.8% improvement in profit, respectively. In contrast, reward shaping, an algorithm that manipulates the reward function, is sensitive to imperfect inputs and leads to a 9.4% decrease in profit. This highlights the importance of selecting an appropriate mechanism when incorporating imperfect inputs.

Figures

Figures reproduced from arXiv: 2505.23355 by the authors.

Figure 1
Figure 1. Schematic diagram of the lettuce greenhouse environment [21]. [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗
Figure 2
Figure 2. Actor-critic architecture [22]. policy is compared to the baseline estimation. At is computed by At = δt + (γλ)δt+1 + · · · + (γλ) T −t+1δT −1 = −vπ(st) + rt + (γλ)rt+1 + · · · + (γλ) T −t vπ(sT ), (3) where GAE trace parameter λ combines multi-step δt to balance the trade-off between bias and variance. 2.3. Interactive RL Backgrounds In baseline RL, human involvement is typically limited to the design of the reward… view at source ↗
Figure 3
Figure 3. Examples of simulated grower giving input. [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (6 more)
Figure 4
Figure 4. Figure 4: Performance of interactive RL algorithms relative to baseline (Vanilla) PPO. [PITH_FULL_IMAGE:figures/full_fig_p020_4.png]
Figure 5
Figure 5. Figure 5: Average setpoints and indoor climate variables over a day. [PITH_FULL_IMAGE:figures/full_fig_p021_5.png]
Figure 6
Figure 6. Figure 6: Impact of the likelihood for policy shaping with precise advice. A: rollout reward, [PITH_FULL_IMAGE:figures/full_fig_p022_6.png]
Figure 7
Figure 7. Figure 7: Impact of the likelihood for policy shaping with precise advice. [PITH_FULL_IMAGE:figures/full_fig_p023_7.png]
Figure 8
Figure 8. Figure 8: Changes in the average loss of F function (control sharing) [PITH_FULL_IMAGE:figures/full_fig_p025_8.png]
Figure 9
Figure 9. Figure 9: Changes in the average loss of πgrower (policy shaping). the selective nor the random strategy is effective in maintaining a low loss for the F function. In contrast, for action advice, the selective strategy consistently outperforms the random strategy in reducing the…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

34 extracted references · 30 canonical work pages

  1. [1]

    Goddek, O

    S. Goddek, O. Körner, K. J. Keesman, M. A. Tester, R. Lefers, L. Fleskens, A. Joyce, E. van Os, A. Gross, R. Leemans, How greenhouse horticulture in arid regions can contribute to climate-resilient and sustainable food se- curity, Global Food Security 38 (2023) 100701

    work page 2023

  2. [2]

    Liao, J.-Y

    P.-A. Liao, J.-Y. Liu, L.-C. Sun, H.-H. Chang, Can the adoption of protected cultivation facilities affect farm sustainability?, Sustainability 12 (23) (2020) 9970

    work page 2020

  3. [3]

    Prakash, P

    P. Prakash, P. Kumar, A. Kar, P. Kishore, A. K. Singh, S. Immanuel, Pro- tected cultivation in maharashtra: determinants of adoption, constraints, and impact, Agricultural Economics Research Review 34 (2) (2021)

    work page 2021

  4. [4]

    Paris, F

    B. Paris, F. Vandorou, A. T. Balafoutis, K. Vaiopoulos, G. Kyriakarakos, D. Manolakos, G. Papadakis, Energy use in greenhouses in the eu: A review recommending energy efficiency measures and renewable energy sources adoption, Applied Sciences 12 (10) (2022) 5150

    work page 2022

  5. [5]

    Social, E

    W. Social, E. Research, Agro & food portal. URLhttps://agrimatie.nl/SectorResultaat.aspx?subpubID=2232& sectorID=2240

    work page

  6. [6]

    Christiaensen, Z

    L. Christiaensen, Z. Rutledge, J. E. Taylor, The future of work in agricul- 29 ture: Some reflections, World Bank Policy Research Working Paper (9193) (2020)

    work page 2020

  7. [7]

    Van Straten, G

    G. Van Straten, G. van Willigenburg, E. van Henten, R. van Ooteghem, Optimal control of greenhouse cultivation, CRC press, 2010

    work page 2010

  8. [8]

    Mahmood, R

    F. Mahmood, R. Govindan, A. Bermak, D. Yang, T. Al-Ansari, Data- driven robust model predictive control for greenhouse temperature control and energy utilisation assessment, Applied Energy 343 (2023) 121190. doi:https://doi.org/10.1016/j.apenergy.2023.121190. URLhttps://www.sciencedirect.com/science/article/pii/ S0306261923005548

    work page arXiv 2023

Show all 34 references
  1. [9]

    Zhang, T

    M. Zhang, T. Yan, W. Wang, X. Jia, J. Wang, J. J. Klemeš, Energy-saving design and control strategy towards modern sustainable greenhouse: A review, Renewable and Sustainable Energy Reviews 164 (2022) 112602. doi:https://doi.org/10.1016/j.rser.2022.112602. URLhttps://www.scien...

    2022

  2. [10]

    Jacobson, P

    B. Jacobson, P. H. Jones, J. Jones, J. Paramore, Real-time greenhouse monitoring and control with an expert system, Com- puters and Electronics in Agriculture 3 (4) (1989) 273–285. doi:https://doi.org/10.1016/0168-1699(89)90018-5. URLhttps://www.sciencedirect.com/science/artic...

    1989 doi

  3. [11]

    Robles Algarín, J

    C. Robles Algarín, J. Callejas Cabarcas, A. Polo Llanos, Low-cost fuzzy logic control for greenhouse environments with web monitoring, Electronics 6 (4) (2017).doi:10.3390/electronics6040071. URLhttps://www.mdpi.com/2079-9292/6/4/71

    2017 doi

  4. [12]

    Pohlheim, A

    H. Pohlheim, A. Heißner, Optimal control of greenhouse climate using ge- netic algorithms, in: Proceedings of the MENDEL, Vol. 96, 1996, pp. 112– 119. 30

    1996

  5. [13]

    J. Lin, Z. Ma, R. Gomez, K. Nakamura, B. He, G. Li, A review on inter- active reinforcement learning from human social feedback, IEEE Access 8 (2020) 120757–120765.doi:10.1109/ACCESS.2020.3006254

    2020

  6. [14]

    Y. Meng, C. Liu, J. Zhao, J. Huang, G. Jing, Stackelberg game-based anti-disturbance control for unmanned surface vessels via integrative rein- forcement learning, Intelligence and Robotics 5 (1) (2025) 88 – 104

    2025

  7. [15]

    Ajagekar, N

    A. Ajagekar, N. S. Mattson, F. You, Energy-efficient ai-based control of semi-closed greenhouses leveraging robust optimization in deep reinforce- ment learning, Advances in Applied Energy 9 (2023) 100119

    2023

  8. [16]

    Morcego, W

    B. Morcego, W. Yin, S. Boersma, E. Van Henten, V. Puig, C. Sun, Rein- forcement learning versus model predictive control on greenhouse climate control, Computers and Electronics in Agriculture 215 (2023) 108372

    2023

  9. [17]

    Zhang, X

    W. Zhang, X. Cao, Y. Yao, Z. An, X. Xiao, D. Luo, Robust model-based reinforcement learning for autonomous greenhouse control, in: Asian Con- ference on Machine Learning, PMLR, 2021, pp. 1208–1223

    2021

  10. [18]

    Griffith, K

    S. Griffith, K. Subramanian, J. Scholz, C. L. Isbell, A. L. Thomaz, Pol- icy shaping: Integrating human feedback with reinforcement learning, Ad- vances in neural information processing systems 26 (2013)

    2013

  11. [19]

    W. B. Knox, P. Stone, Reinforcement learning from simultaneous human and mdp reward, in: AAMAS, Vol. 1004, Valencia, 2012, pp. 475–482

    2012

  12. [20]

    Moreira, J

    I. Moreira, J. Rivas, F. Cruz, R. Dazeley, A. Ayala, B. Fernandes, Deep reinforcement learning with interactive feedback in a human–robot envi- ronment, Applied Sciences 10 (16) (2020) 5574

    2020

  13. [21]

    Van Henten, Greenhouse climate management: an optimal control ap- proach, Wageningen University and Research, 1994

    E. Van Henten, Greenhouse climate management: an optimal control ap- proach, Wageningen University and Research, 1994

    1994

  14. [22]

    Schulman, F

    J. Schulman, F. Wolski, P. Dhariwal, A. Radford, O. Klimov, Proximal policy optimization algorithms, arXiv preprint arXiv:1707.06347 (2017). 31

    2017 arXiv

  15. [23]

    Raffin, A

    A. Raffin, A. Hill, A. Gleave, A. Kanervisto, M. Ernestus, N. Dormann, Stable-baselines3: Reliable reinforcement learning implementations, Jour- nal of Machine Learning Research 22 (268) (2021) 1–8

    2021

  16. [24]

    Arzate Cruz, T

    C. Arzate Cruz, T. Igarashi, A survey on interactive reinforcement learning: Design principles and open challenges, in: Proceedings of the 2020 ACM designing interactive systems conference, 2020, pp. 1195–1209

    2020

  17. [25]

    W. B. Knox, P. Stone, Combining manual feedback with subsequent mdp reward signals for reinforcement learning, in: AAMAS, Vol. 10, 2010, pp. 5–12

    2010

  18. [26]

    Cederborg, I

    T. Cederborg, I. Grover, C. L. Isbell Jr, A. L. Thomaz, Policy shaping with human teachers, in: IJCAI, 2015, pp. 3366–3372

    2015

  19. [27]

    W. B. Knox, P. Stone, Tamer: Training an agent manually via evaluative reinforcement, in: 2008 7th IEEE international conference on development and learning, IEEE, 2008, pp. 292–297

    2008

  20. [28]

    Bignold, F

    A. Bignold, F. Cruz, R. Dazeley, P. Vamplew, C. Foale, An evaluation methodology for interactive reinforcement learning with simulated users, Biomimetics 6 (1) (2021) 13

    2021

  21. [29]

    Cacciarelli, M

    D. Cacciarelli, M. Kulahci, Active learning for data streams: a survey, Machine Learning 113 (1) (2024) 185–239

    2024

  22. [30]

    Zhang, K

    J. Zhang, K. Cho, Query-efficient imitation learning for end-to-end au- tonomous driving, arXiv preprint arXiv:1605.06450 (2016)

    2016 arXiv

  23. [31]

    URLhttps://www.knmi.nl/nederland-nu/klimatologie/uurgegevens

    KNMI, Knmi - hourly weather data in the netherlands. URLhttps://www.knmi.nl/nederland-nu/klimatologie/uurgegevens

  24. [32]

    X. Lan, P. Tans, K. Thoning, Trends in globally-averaged co2 determined from noaa global monitoring laboratory measurements.doi:https://doi. org/10.15138/9N0H-ZH07. 32

    doi

  25. [33]

    Huang, R

    S. Huang, R. F. J. Dossa, A. Raffin, A. Kanervisto, W. Wang, The 37 im- plementation details of proximal policy optimization, in: ICLR Blog Track, 2022, https://iclr-blog-track.github.io/2022/03/25/ppo-implementation- details/. URLhttps://iclr-blog-track.github.io/2022/03/25/ ...

    2022

  26. [34]

    T. Kobayashi, Intentionally-underestimated value function at terminal state for temporal-difference learning with mis-designed reward, Results in Control and Optimization 18 (2025) 100530. 33

    2025

Pith tools

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