1. the complete title of one (or more) paper(s) published in the open literature describing the work that the author claims describes a human-competitive result; Analyzing Interpretable Visual Control Policy Search and Synthesis 2. the name, complete physical mailing address, e-mail address, and phone number of EACH author of EACH paper(s); Christina Berghegger 3 rue du Puits Vert, 31000 Toulouse, France christina.berghegger@ut-capitole.fr +33 7 76 24 18 57 Camilo De La Torre 14 rue de Negrenys, 31200 Toulouse, France dlt.camilo@gmail.com +33 7 82 99 23 22 Giorgia Nadizar 6 Rue des Gestes, 31000 Toulouse, France giorgia.nadizar@ut-capitole.fr +33 6 10 34 86 90 Yuri Lavinas 6 rue Félix Debax, 31700 Blagnac, France yuri.lavinas@ut-capitole.fr +33 6 01 30 09 24 David Simoncini 37 Chemin de la Butte, 31400 Toulouse, France david.simoncini@ut-capitole.fr +33 6 51 31 99 95 Hervé Luga 4 bis rue marcel pagnol, 31100 Toulouse, France herve.luga@irit.fr +33 6 89 44 58 27 Dennis G. Wilson 2840 Chem. de Marcenac, 82600 Mas-Grenier, France dennis.wilson@isae-supaero.fr +33 6 23 54 21 13 Sylvain Cussat-Blanc 50 rue des Peupliers, 31700 Beauzelle, France sylvain.cussat-blanc@ut-capitole.fr +33 6 88 70 62 00 3. the name of the corresponding author (i.e., the author to whom notices will be sent concerning the competition); Christina Berghegger 4. the abstract of the paper(s); Vision-based decision-making is relevant to many domains, including safety-critical ones where transparency matters as much as performance. Therefore, automating sequential decision making in such settings requires approaches that balance effectiveness with interpretability. While deep reinforcement learning techniques based on artificial neural networks have achieved strong performance, their black-box nature typically necessitates post hoc explainability analyses. To address this limitation, we propose an approach based on Graph-based Genetic Programming (GGP) that generates policies in the form of computer code, which is fully observable and inherently interpretable. To improve both performance and robustness, we expose GGP-based visual control policies to multiple representative conditions during optimization, mitigating convergence to strong local optima and fragility under changing conditions. Finally, to gain insights into the evolutionary search dynamics, we employ Search Trajectory Networks, an analytical and visualization tool for studying optimization behavior. Our results demonstrate that the resulting policies approach human-level performance and empirically confirm the presence of strong local optima acting as attractors during evolution, providing new insights into the behavior and potential limitations of interpretable evolutionary policy search approaches. 5. a list containing one or more of the four letters (A, B, C, or D) that correspond to the criteria (see above) that the author claims that the work satisfies; B, C 6. a statement stating why the result satisfies the criteria that the contestant claims (see examples of statements of human-competitiveness as a guide to aid in constructing this part of the submission); B: The result was published in the ACM journal Transactions on Evolutionary Learning and Optimization C: With our approach, we reach results comparable to both state-of-the-art methods and human performance on three games of the Atari benchmark. Human results are often reported as average, while reinforcement learning (RL) results are generally reported as the best performance. For the three games our approach is tested on, results are as follow: - Pong: The average human results is 14.6, our average result is 19.47, and we reach the same best score as RL, which is 21 (maximum points) - Freeway: The average human result is 29.6, our average result is 26.09, the best score reached with RL is 34, while we reach up to 28. - Bowling: The average human result is 160.7, our average result is 172.2, the best score reached with RL surpasses 200, while we reach up to 223 The current state-of-the-art methods are mainly achieved with black-box models. In contrast, our results prove both high performance and high interpretability, the latter being hardly achievable with deep learning approaches. Interpretability can provide insights on the strategies to humans to improve their results. 7. a full citation of the paper (that is, author names; title, publication date; name of journal, conference, or book in which article appeared; name of editors, if applicable, of the journal or edited book; publisher name; publisher city; page numbers, if applicable); Christina Berghegger, Camilo De La Torre, Giorgia Nadizar, Yuri Lavinas, David Simoncini, Hervé Luga, Dennis G. Wilson, and Sylvain Cussat-Blanc. 2026. Analyzing Interpretable Visual Control Policy Search and Synthesis. ACM Trans. Evol. Learn. Optim. Just Accepted (April 2026). https://doi.org/10.1145/3799432 8. a statement either that "any prize money, if any, is to be divided equally among the co-authors" OR a specific percentage breakdown as to how the prize money, if any, is to be divided among the co-authors; Any prize money, if any, is to be divided equally among the co-authors. 9. a statement stating why the authors expect that their entry would be the "best," and With this contribution, we aim to demonstrate that it is possible to apply white-box and inherently interpretable artificial intelligence methods without sacrificing performance. Using our approach, we achieved results comparable to both state-of-the-art methods and human performance, while maintaining full insight into the internal structure of the generated policy. With expert knowledge, the behavior of our agent could be fully explained. In light of the increasing regulation surrounding the application of AI, particularly in safety-critical domains, the inherent interpretability of our approach makes it highly promising for future applications. 10. An indication of the general type of genetic or evolutionary computation used, such as GA (genetic algorithms), GP (genetic programming), ES (evolution strategies), EP (evolutionary programming), LCS (learning classifier systems), GI (genetic improvement), GE (grammatical evolution), GEP (gene expression programming), DE (differential evolution), etc. GA, GP, CGP 11. The date of publication of each paper. If the date of publication is not on or before the deadline for submission, but instead, the paper has been unconditionally accepted for publication and is “in press” by the deadline for this competition, the entry must include a copy of the documentation establishing that the paper meets the "in press" requirement. Accepted for publication at TELO on 23 February 2026 and available online in its final version since 11 April 2026: https://dl.acm.org/doi/abs/10.1145/3799432. The paper is "in press" by the competition deadline. The acceptance notification email and the camera-ready version with copyright statement are attached.