Pith. sign in

REVIEW 4 major objections 5 minor 27 references

Explainable Reinforcement Learning for Formula One Race Strategy

T0 review · 4 major / 5 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims a lap-by-lap reinforcement learning agent can outperform Mercedes' own strategy model in the team's race simulator.

desk verdict A genuine applied-RL contribution with real external baselines, but the headline number is selected on the test set and the statistical reporting is too thin to back the central claim. read the letter →

arxiv 2501.04068 v1 pith:ZHWCJU6R submitted 2025-01-07 cs.LG cs.AI

Devin Thomas , Junqi Jiang , Avinash Kori , Aaron Russo , Steffen Winkler , Stuart Sale , Joseph McMillan , Francesco Belardinelli
show 1 more author
Antonio Rago
This is my paper · ORCID
classification cs.LGcs.AI
keywords reinforcementlearningexplainableAIFormulaOneracestrategydeeprecurrentQ-networkMonteCarlosimulationtyredegradationcounterfactualexplanations
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 claims that a reinforcement learning agent, RSRL, can decide lap by lap whether to pit and which tyre compound to fit, and that this agent performs better than two established alternatives in a Formula One race simulator. On the 2023 Bahrain Grand Prix, with a car whose pace corresponds to an expected finish of P5.5, RSRL averages P5.33 over 1920 simulations, compared with P5.63 for the Fixed Strategy model and P5.86 for the state-of-the-art model supplied by the Mercedes team. The authors argue that the improvement is meaningful because the simulator's stochasticity and the car's pace make consistently finishing near the front impossible. They also report that training on one or two tracks produces excellent results on those tracks but transfers poorly, while training on nine tracks transfers better to unseen tracks at a cost on the training tracks. The paper adds explanation tools to make the agent's strategies inspectable.

What carries the argument

The central object is the RSRL agent, a deep recurrent Q-network (DRQN) that uses a recurrent neural network to incorporate information from previous laps when predicting Q-values, which is what lets it handle partial observability such as whether it is catching or losing ground to the car ahead. The agent is connected to the simulator through an abstraction layer: custom UnifiedRaceState and UnifiedRaceStrategy classes with a translator convert the simulator's proprietary data into a fixed state representation, so the same agent can be trained on the Monte Carlo simulator and deployed on other data sources. The reward function is terminal-position based, giving 100 times the F1 points for the eventual finish, with a penalty for extra pit stops and a large penalty for illegal actions, which lets the agent learn purely from race outcomes without intermediate reward shaping.

What would settle it

Record the finishing position of each of the 1920 simulations and run a permutation or bootstrap test comparing RSRL's mean P5.33 against Mercedes SOTA's P5.86; if the difference is not statistically significant, the central performance claim reduces to noise. A complementary check is to replay RSRL's learned strategies in a different simulator or on real historical race data and compare their finishing positions with the baselines.

Watch

Extended reading notes

Core claim

RSRL is a deep recurrent Q-network that treats each lap of a race as a decision step. At each step the available actions are no pit stop, pit for soft tyres, pit for medium tyres, or pit for hard tyres; the reward is mostly terminal, equal to 100 times the Formula One points for the final finishing position, with penalties for extra pit stops and heavy penalties for invalid actions. Trained inside the team's black-box Monte Carlo simulator, the agent learns to select strategies that lift a car with P5.5 expected pace to an average P5.33 finish on the Bahrain 2023 test race, ahead of both baselines. The paper's generalisability study shows that the number of training tracks controls a trade-off between performance on known circuits and robustness on unseen circuits.

Load-bearing premise

The evaluation assumes the Mercedes Monte Carlo simulator faithfully represents real Formula One racing, including tyre degradation, traffic, and safety-car behaviour, so a strategy that wins in simulation will also win on track.

Editorial extensions

If this is right

  • A team using RSRL could choose strategy online, one lap at a time, instead of committing to a small set of pre-computed candidate strategies and evaluating them with Monte Carlo simulation.
  • The P5.33 average over 1920 simulations is better than the P5.63 of the Fixed Strategy baseline and the P5.86 of the Mercedes SOTA model, so the agent extracts more from a car with P5.5 expected pace.
  • The generalisability study implies a practical rule: train on a small number of tracks for peak performance on a known calendar, or train on many tracks for robustness to unseen circuits, accepting a performance loss on the training tracks.
  • The XAI components mean a strategist can inspect why a particular lap's decision was made, test what would change the decision, and retain a decision-tree-level view of the whole policy.
  • The abstraction layer between the agent and the simulator means the same trained model can be deployed with live race data or a driver-in-the-loop simulator without retraining the core decision logic.

Reading between the lines

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

  • A natural next step the paper leaves implicit is to report the variance across the 1920 simulation seeds; without a confidence interval or significance test, the 5.33 vs 5.63 vs 5.86 ordering could be within sampling noise.
  • If the result holds, the same DRQN-plus-abstraction design should transfer to other tyre-limited racing series such as Formula E or endurance racing, because their strategy problems share the same action structure of when to pit and which compound to fit.
  • The fact that RSRL's counterfactual for an early soft-tyre stop lands on lap 14, the lap where the actual 2023 Bahrain winner pitted, suggests the learned policy may track real-world optimal behaviour; this could be checked by comparing RSRL's strategies with the actual winning strategies of other grands prix.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 5 minor

Summary. The paper introduces RSRL, a DRQN-based reinforcement learning agent for Formula One race strategy (tyre compound selection and pit-stop timing), trained inside a proprietary Monte Carlo race simulator provided by Mercedes-AMG PETRONAS F1 Team. The central empirical claim is that RSRL achieves an average finishing position of P5.33 on the 2023 Bahrain Grand Prix over 1920 simulations, outperforming a Fixed Strategy baseline (P5.63) and Mercedes' SOTA model (P5.86). The paper also presents a generalisability study across 14 tracks for models trained on 1, 2 and 9 tracks, and supplements the policy with three XAI techniques: TimeSHAP feature importance, VIPER decision-tree surrogates, and decision-tree counterfactuals.

Significance. If the reported results are statistically sound, the paper would be a useful applied contribution: it demonstrates a portable RL architecture for real-time race strategy, evaluates against an industry black-box baseline, and combines policy learning with several XAI tools in a domain where such explanations are directly relevant to human strategists. The system-architecture abstraction and the generalisability experiment across multiple tracks are constructive strengths. However, the headline advantage is currently supported only by a point estimate obtained after testing several model versions on the same race, with no uncertainty quantification; the result is not yet established as a robust property of the method.

major comments (4)
  1. [§4.1] The main comparison is not statistically controlled because the test race is used for model selection. The sentence 'Through testing, we found that the best-performing version of RSRL for the 2023 Bahrain Grand Prix achieved an average finishing position of P5.33' indicates that P5.33 is the best of several evaluated variants, whereas the Fixed Strategy and Mercedes SOTA baselines are single fixed models. Table 3 shows other RSRL instances (RSRL-1, RSRL-2, RSRL-9) finishing Bahrain at 9.62, 6.47 and 6.45, all worse than both baselines, so the reported advantage is an outlier rather than a robust property. The manuscript should select the model on a validation split, or report all evaluated variants with a multiple-comparison correction, and then evaluate the selected model on a held-out race.
  2. [§4.1 and §4.2] All headline results are point estimates without any measure of dispersion or uncertainty. The 1920-simulation comparison in §4.1 and the approximately 900-simulation-per-track comparisons in Table 3 report only average finishing positions; no standard deviations, confidence intervals, or significance tests are given. Given the stochastic simulator, a 0.30-position gap (P5.33 vs P5.63) or a 0.53-position gap (P5.33 vs P5.86) cannot be distinguished from simulation noise. The same issue affects the generalisability claims, e.g. 'RSRL-1 and RSRL-2 finish on average 0.93 and 1.77 positions ahead of the SOTA model respectively when tested on their training tracks.' The authors should provide bootstrap confidence intervals or paired permutation tests, and report per-model distributions rather than only means.
  3. [§4.3] The XAI fidelity evaluation does not support the stated conclusions. For TimeSHAP, the MAE of 124.39 is normalized by the maximum terminal reward of 2500 to claim '5% normalised MAE', but the intermediate rewards are mostly 1 or -10, so this normalization is not meaningful unless the error is measured in reward units or against a baseline such as constant/random attribution. For VIPER, the 0.926 accuracy and 0.910 F1-score are computed on 100 simulations with a strongly imbalanced label distribution (4699 no-pit true labels in Table 4), so per-class precision and recall are needed to assess fidelity. The counterfactual 'fidelity' measure, which counts the number of feature changes, is a proximity measure and does not directly measure how closely the counterfactual follows the model's decision boundary; this should be clarified or replaced with a decision-boundary-based check.
  4. [§3.1] The reward function in the displayed equation does not match the prose description. The text says a penalty of -10 is given for 'extraneous pitstops beyond the first valid pitstop,' but the equation applies -10 whenever a_t is a pit action and s_vf is true, without tracking whether a valid pitstop has already occurred. Since this reward affects the learned policy and hence the performance claim, the definition should be corrected or clarified, and the implementation should be checked against the intended semantics.
minor comments (5)
  1. [§3.1 / Table 1] Table 1 lists the ranges of Tyre Degradation, Gap Ahead, Gap Behind, and Gap to Leader as '?', leaving the scaling bounds unspecified; the authors should provide the actual ranges used in the custom scaling functions.
  2. [§4.3 / Figure 4] The text says 'The Lap Number plot is the most influential,' but the state space in Table 1 does not contain a Lap Number feature; the relevant feature appears to be Race Progress. The caption and the text should be made consistent.
  3. [§5] There is a typo in 'we would like to to improve strategy predictions'; the duplicated 'to' should be removed.
  4. [§2 / Reference [7]] Reference [7] is cited as 'Molnar and Dandl's [7]' but the reference list entry is authored by Dandl, Molnar, Binder, and Bischl; the in-text citation and reference entry should be aligned.
  5. [§4.3] The statement 'there are considerably more no pit decisions being taken through the 100 simulations' is imprecise; it should say that the simulated trajectories contain many more no-pit timesteps than pit timesteps, which explains the class imbalance in the confusion matrix.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the central claim is an empirical comparison against external baselines, not a derivation from fitted parameters or self-citations.

full rationale

The paper's main claim—that RSRL achieves P5.33 on the 2023 Bahrain Grand Prix, beating Fixed Strategy at P5.63 and Mercedes SOTA at P5.86—is an empirical result obtained from a proprietary Monte Carlo simulator against external baselines. The RL agent is trained with a reward function and then evaluated; no parameter is fitted to the target finishing position by construction. The phrase 'Through testing, we found that the best-performing version of RSRL for the 2023 Bahrain Grand Prix achieved an average finishing position of P5.33' indicates model selection on the test race, which is a statistical robustness concern (the reported number may be optimistically selected, especially since Table 3 shows other RSRL variants finishing worse), but it is not circular reasoning: P5.33 is a simulation outcome, not an algebraic consequence of the baselines or of the model definition. The XAI evaluations compare surrogate models and SHAP values to the same RSRL model, which is standard practice for measuring explanation fidelity, not circularity of the main result. There is no load-bearing self-citation, no imported uniqueness theorem, and no ansatz smuggled in via prior work. The derivation chain is self-contained relative to the simulator, and the identified selection-bias issue should be treated as a correctness/statistical limitation rather than circularity.

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

The paper's claims rest on a proprietary simulator, hand-tuned reward coefficients, manually estimated feature ranges, and a checkpoint selected on the test race. It introduces no new physical or conceptual entities.

free parameters (4)
  • Reward coefficients (step +1, extra pit stop -10, invalid action -1000, terminal 100x points) = chosen by hand
    The relative scale of step rewards, penalties, and terminal rewards is hand-set and directly shapes the learned policy; no sensitivity analysis is provided.
  • Feature scaling ranges for state normalization = estimated from pilot simulations
    Custom linear scaling functions are built from value ranges observed in simulations; these ranges are a form of data-dependent preprocessing.
  • DRQN hyperparameters (epsilon decay 0.999, gamma 0.99, learning rate 0.001, replay buffer 1000) = final values in Table 2
    Hyperparameters are manually tuned; the paper does not describe a validation procedure.
  • Best model checkpoint selection = chosen on test race
    'Through testing, we found the best-performing version of RSRL' implies model selection on the evaluation race, an optimistic bias.
assumptions (4)
  • domain assumption The Monte Carlo simulator accurately models F1 race dynamics and car pace.
    Section 3.2: the simulator is a black box; all training and evaluation use it, so its fidelity determines whether results transfer.
  • domain assumption The driver's pace is equivalent to an expected finishing position of P5.5.
    Section 4.1: the car is set to this pace; if the calibration is wrong, the meaning of the finishing positions shifts.
  • domain assumption The four-action set (no pit, pit soft/medium/hard) covers all relevant strategy choices.
    Section 3.1: only these actions are available; the F1 regulations require at least one pit stop to a different compound, but other actions like fuel modes are ignored.
  • domain assumption DRQN converges to a good policy under the given reward.
    Section 3.2: the paper assumes the trained DRQN is a reasonable optimizer; no convergence guarantees.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Explainable Reinforcement Learning for Formula One Race Strategy." pith.science (2026). https://pith.science/paper/ZHWCJU6R

@misc{pith2026250104068,
  author       = {Pith},
  title        = {Pith review of: Explainable Reinforcement Learning for Formula One Race Strategy},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZHWCJU6R}},
  note         = {Machine review of arXiv:2501.04068}
}
read the original abstract

In Formula One, teams compete to develop their cars and achieve the highest possible finishing position in each race. During a race, however, teams are unable to alter the car, so they must improve their cars' finishing positions via race strategy, i.e. optimising their selection of which tyre compounds to put on the car and when to do so. In this work, we introduce a reinforcement learning model, RSRL (Race Strategy Reinforcement Learning), to control race strategies in simulations, offering a faster alternative to the industry standard of hard-coded and Monte Carlo-based race strategies. Controlling cars with a pace equating to an expected finishing position of P5.5 (where P1 represents first place and P20 is last place), RSRL achieves an average finishing position of P5.33 on our test race, the 2023 Bahrain Grand Prix, outperforming the best baseline of P5.63. We then demonstrate, in a generalisability study, how performance for one track or multiple tracks can be prioritised via training. Further, we supplement model predictions with feature importance, decision tree-based surrogate models, and decision tree counterfactuals towards improving user trust in the model. Finally, we provide illustrations which exemplify our approach in real-world situations, drawing parallels between simulations and reality.

Figures

Figures reproduced from arXiv: 2501.04068 by the authors.

Figure 1
Figure 1. The system architecture implemented utilises abstraction to provide future flexibility, extensibility and portability. It [PITH_FULL_IMAGE:figures/full_fig_p004_1.png] view at source ↗
Figure 2
Figure 2. Average finishing positions for Mercedes’ SOTA model, RSRL, and the Fixed Strategy model. Prix.8 Mercedes’ SOTA model is based on a probabilistic model using heuristics in its decision-making, but is treated as a black￾box. Through testing, we found that the best-performing version of RSRL for the 2023 Bahrain Grand Prix achieved an average finishing position of P5.33 over 1920 simulations, whilst the Fixed Strategy… view at source ↗
Figure 3
Figure 3. The most common tyre strategies generated by each generalisability model, [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figures from the paper (3 more)
Figure 4
Figure 4. Figure 4: The feature importance for RSRL on lap 10 of the [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]
Figure 6
Figure 6. Figure 6: Accuracy of VIPER decision tree testing with differ [PITH_FULL_IMAGE:figures/full_fig_p007_6.png]
Figure 5
Figure 5. Figure 5: Accuracy during training and testing of 50 iterations of VIPER decision trees. Each iteration covers all previous data points queried from the oracle (the best RSRL model), up to 26,714 in iteration 50. VIPER Decision Tree Surrogate Model. The fidelity of the VIPER dec…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 15 canonical work pages

  1. [1]

    Ali El-Sappagh, Khan Muhammad, Jose Maria Alonso-Moral, Roberto Confalonieri, Riccardo Guidotti, Javier Del Ser, Natalia Díaz Rodríguez, and Francisco Herrera

    Sajid Ali, Tamer Abuhmed, Shaker H. Ali El-Sappagh, Khan Muhammad, Jose Maria Alonso-Moral, Roberto Confalonieri, Riccardo Guidotti, Javier Del Ser, Natalia Díaz Rodríguez, and Francisco Herrera. 2023. Explainable Artificial Intelligence (XAI): What we know and what is left to attain Trustworthy Artificial Intelligence. Inf. Fusion 99 (2023), 101805. http...

    doi:10.1016/j.inffus.2023 2023

  2. [2]

    Osbert Bastani, Yewen Pu, and Armando Solar-Lezama. 2018. Verifiable Reinforcement Learning via Policy Extraction. In NeurIPS. 2499–2509. https:// proceedings.neurips.cc/paper/2018/hash/e6d8545daa42d5ced125a4bf747b3688- Abstract.html

    work page 2018

  3. [3]

    João Bento, Pedro Saleiro, André Ferreira Cruz, Mário A. T. Figueiredo, and Pedro Bizarro. 2021. TimeSHAP: Explaining Recurrent Models through Sequence Perturbations. In KDD. 2565–2573. https://doi.org/10.1145/3447548.3467166

    arXiv 2021

  4. [4]

    Christopher Berner, Greg Brockman, Brooke Chan, Vicki Cheung, Przemys- law Debiak, Christy Dennison, David Farhi, Quirin Fischer, Shariq Hashme, Christopher Hesse, Rafal Józefowicz, Scott Gray, Catherine Olsson, Jakub Pa- chocki, Michael Petrov, Henrique Pondé de Oliveira Pinto, Jonathan Raiman, Tim Salimans, Jeremy Schlatter, Jonas Schneider, Szymon Sido...

    arXiv 2019

  5. [5]

    Max Boettinger and David Klotz. 2023. Mastering Nordschleife - A com- prehensive race simulation for AI strategy decision-making in motorsports. CoRR abs/2306.16088 (2023). https://doi.org/10.48550/ARXIV.2306.16088 arXiv:2306.16088

    work page Pith review arXiv doi:10.48550/arxiv.2306.16088 2023

  6. [6]

    Carreira-Perpiñán and Suryabhan Singh Hada

    Miguel Á. Carreira-Perpiñán and Suryabhan Singh Hada. 2021. Counterfactual Explanations for Oblique Decision Trees: Exact, Efficient Algorithms. In AAAI. 6903–6911. https://doi.org/10.1609/AAAI.V35I8.16851

    work page doi:10.1609/aaai.v35i8.16851 2021

  7. [7]

    Susanne Dandl, Christoph Molnar, Martin Binder, and Bernd Bischl. 2020. Multi- Objective Counterfactual Explanations. In PPSN. 448–469. https://doi.org/10. 1007/978-3-030-58112-1_31

    work page 2020

  8. [8]

    Samuel Greydanus, Anurag Koul, Jonathan Dodge, and Alan Fern. 2018. Visualiz- ing and Understanding Atari Agents. In PMLR. 1792–1801. https://proceedings. mlr.press/v80/greydanus18a.html

    work page 2018

Show all 27 references
  1. [9]

    Hausknecht and Peter Stone

    Matthew J. Hausknecht and Peter Stone. 2015. Deep Recurrent Q-Learning for Partially Observable MDPs. In AAAI Fall Symposia. 29–37. http://www.aaai.org/ ocs/index.php/FSS/FSS15/paper/view/11673

    2015

  2. [10]

    Bradley Hayes and Julie A. Shah. 2017. Improving Robot Controller Transparency Through Autonomous Policy Explanation. In HRI. 303–312. https://doi.org/10. 1145/2909824.3020233

    2017

  3. [11]

    Alexander Heilmeier, André Thomaser, Michael Graf, and Johannes Betz. 2020. Virtual Strategy Engineer: Using Artificial Neural Networks for Making Race Strategy Decisions in Circuit Motorsport. Applied Sciences (2020). https://api. semanticscholar.org/CorpusID:228907155

    2020

  4. [12]

    Carrasco Heine and Charles Thraves

    Oscar F. Carrasco Heine and Charles Thraves. 2023. On the optimization of pit stop strategies via dynamic programming. Central Eur. J. Oper. Res. 31, 1 (2023), 239–268. https://doi.org/10.1007/S10100-022-00806-4

    2023 doi

  5. [13]

    Jalali, Bernhard Haslhofer, Simone Kriglstein, and Andreas Rauber

    Anahid N. Jalali, Bernhard Haslhofer, Simone Kriglstein, and Andreas Rauber

  6. [14]

    Xuze Liu and Abbas Fotouhi. 2020. Formula-E race strategy development using artificial neural networks and Monte Carlo tree search. Neural Comput. Appl. 32, 18 (2020), 15191–15207. https://doi.org/10.1007/S00521-020-04871-1

    2020 doi

  7. [15]

    Xuze Liu, Abbas Fotouhi, and Daniel J. Auger. 2021. Formula-E race strategy development using distributed policy gradient reinforcement learning. Knowl. Based Syst. 216 (2021), 106781. https://doi.org/10.1016/J.KNOSYS.2021.106781

    2021

  8. [16]

    Lundberg and Su-In Lee

    Scott M. Lundberg and Su-In Lee. 2017. A Unified Approach to Interpreting Model Predictions. In NeurIPS. 4765–4774. https://proceedings.neurips.cc/paper/ 2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html

    2017

  9. [17]

    Stephanie Milani, Nicholay Topin, Manuela Veloso, and Fei Fang. 2024. Explain- able Reinforcement Learning: A Survey and Comparative Review. ACM Comput. Surv. 56, 7 (2024), 168:1–168:36. https://doi.org/10.1145/3616864

    2024 doi

  10. [18]

    Riedmiller

    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Alex Graves, Ioannis Antonoglou, Daan Wierstra, and Martin A. Riedmiller. 2013. Playing Atari with Deep Reinforcement Learning. CoRR abs/1312.5602 (2013). arXiv:1312.5602 http://arxiv.org/abs/1312.5602

    2013 arXiv

  11. [19]

    Rusu, Joel Veness, Marc G

    Volodymyr Mnih, Koray Kavukcuoglu, David Silver, Andrei A. Rusu, Joel Veness, Marc G. Bellemare, Alex Graves, Martin A. Riedmiller, Andreas Fidjeland, Georg Ostrovski, Stig Petersen, Charles Beattie, Amir Sadik, Ioannis Antonoglou, Helen King, Dharshan Kumaran, Daan Wierstra, ...

    2015 doi

  12. [20]

    Gordon, and Drew Bagnell

    Stéphane Ross, Geoffrey J. Gordon, and Drew Bagnell. 2011. A Reduction of Imitation Learning and Structured Prediction to No-Regret Online Learning. In AISTATS. 627–635. http://proceedings.mlr.press/v15/ross11a/ross11a.pdf

    2011

  13. [21]

    Lloyd S Shapley. 1953. A Value for n-Person Games. InContributions to the Theory of Games II, Harold W. Kuhn and Albert W. Tucker (Eds.). Princeton University Press, Princeton, 307–317

    1953

  14. [22]

    David Silver, Aja Huang, Chris J. Maddison, Arthur Guez, Laurent Sifre, George van den Driessche, Julian Schrittwieser, Ioannis Antonoglou, Vedavyas Pan- neershelvam, Marc Lanctot, Sander Dieleman, Dominik Grewe, John Nham, Nal Kalchbrenner, Ilya Sutskever, Timothy P. Lillicra...

    2016 doi

  15. [23]

    Harm van Seijen, Mehdi Fatemi, Romain Laroche, Joshua Romoff, Tavian Barnes, and Jeffrey Tsang. 2017. Hybrid Reward Architecture for Reinforce- ment Learning. In NIPS. 5392–5402. https://proceedings.neurips.cc/paper/2017/ hash/1264a061d82a2edae1574b07249800d6-Abstract.html

    2017

  16. [24]

    Abhinav Verma, Vijayaraghavan Murali, Rishabh Singh, Pushmeet Kohli, and Swarat Chaudhuri. 2018. Programmatically Interpretable Reinforcement Learning. In ICML. 5052–5061. http://proceedings.mlr.press/v80/verma18a.html

    2018

  17. [25]

    Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung Chung, David H

    Oriol Vinyals, Igor Babuschkin, Wojciech M. Czarnecki, Michaël Mathieu, Andrew Dudzik, Junyoung Chung, David H. Choi, Richard Powell, Timo Ewalds, Petko Georgiev, Junhyuk Oh, Dan Horgan, Manuel Kroiss, Ivo Danihelka, Aja Huang, Laurent Sifre, Trevor Cai, John P. Agapiou, Max J...

    2019

  18. [26]

    Wurman, Samuel Barrett, Kenta Kawamoto, James MacGlashan, Kaushik Subramanian, Thomas J

    Peter R. Wurman, Samuel Barrett, Kenta Kawamoto, James MacGlashan, Kaushik Subramanian, Thomas J. Walsh, Roberto Capobianco, Alisa Devlic, Franziska Eckert, Florian Fuchs, Leilani Gilpin, Piyush Khandelwal, Varun Raj Kompella, HaoChih Lin, Patrick MacAlpine, Declan Oller, Taku...

    2022

  19. [2023]

    CoRR abs/2309.11987 (2023)

    Predictability and Comprehensibility in Post-Hoc XAI Methods: A User- Centered Analysis. CoRR abs/2309.11987 (2023). https://doi.org/10.48550/ARXIV. 2309.11987 arXiv:2309.11987

    2023 doi arXiv

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.