[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-detail-84975-en":3,"doc-seo-84975-105":29,"detail-sidebar-cat-0-en-105":91},{"code":4,"msg":5,"data":6},0,"success",{"doc_id":7,"user_id":8,"nickname":9,"user_avatar":10,"doc_module":4,"category_id":11,"category_name":12,"doc_title":13,"doc_description":14,"doc_content":15,"file_id":16,"file_url":17,"file_type":18,"file_size":19,"view_count":20,"is_deleted":4,"is_public":20,"is_downloadable":20,"audit_status":20,"page_count":21,"language":22,"language_code":23,"site_id":24,"html_lang":23,"table_of_contents":25,"faqs":26,"seo_title":13,"seo_description":14,"update_tm":27,"read_time":28},84975,7971461740909,"Levi","https://ap-avatar.wpscdn.com/davatar_155a257f0dc6eb9ab79c44ca47cae57d",8,"Research & Report","ORCAID Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies","Explainability is a major challenge in reinforcement learning when policies must be both trustworthy and understandable. ORCAID extracts interpretable, rule-based surrogate policies from deep RL agents in mixed continuous–discrete environments with continuous action spaces. It uses an efficient oblique decision tree training procedure that partitions the state space with hyperplanes and fits local linear models. A three-stage split search and subsequent backward elimination reduce complexity, while merging adjacent leaves yields concise rules. Across multiple environments, ORCAID preserves strong performance with few parameters and can improve the original policy.","ORCAID: Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies  \nIgnacio D. Lopez-Miguel  \nTU Wien Vienna, Austria  \n[ignacio.lopez@tuwien.ac.at](ignacio.lopez@tuwien.ac.at)  \nThomas Eiter  \nTU Wien Vienna, Austria  \n[thomas.eiter@tuwien.ac.at](thomas.eiter@tuwien.ac.at)  \nEzio Bartocci  \nTU Wien Vienna, Austria  \n[ezio.bartocci@tuwien.ac.at](ezio.bartocci@tuwien.ac.at)  \nMartin Tappler  \nTU Wien Vienna, Austria  \n[martin.tappler@tuwien.ac.at](martin.tappler@tuwien.ac.at)  \narXiv :2607 .07235v 1 [ cs .LG] 8 Jul 2026  \nAbstract  \nExplainability remains a key issue in reinforcement learning (RL) . Distilling an interpretable policy from an agent trained in a complex environment is particularly challenging when the action space is continuous. We introduce ORCAID, a novel method for extracting interpretable rule-based policies from RL agents operating in mixed continuous-discrete environments with continuous action spaces. Our main contribution is an efficient oblique decision tree training algorithm that partitions the state space by hyperplanes and fits local linear models. The key idea lies ina three-stage split search: efficient random initialization, local refinement, and backward elimination. Finally, adjacent leaves are merged to yield a concise set of interpretable rules describing a given deep RL policy. We evaluate ORCAID across multiple RL environments, demonstrating that the extracted rule-based policies maintain strong performance with a low number of parameters and can even be used to improve the performance of the original deep RL policy.  \n1 Introduction  \nReinforcement learning (RL) [42] builds decision-making policies by trial and error to optimize rewards associated with a control task. It substantially advanced in the last decade by the integration of deep neural networks into training; such deep RL approaches have managed to learn to play computer games at a human level [24] and reached superhuman performance in complex board games, like Go [40] . Deep RL has also ventured into areas where reward-based task specifications are simpler than alternative formulations. Recent applications include fine-tuning LLMs [46], learning policies for control problems in the natural sciences [8], and use in several autonomous systems [14] .  \nDespite this success, the application of deep RL is hampered by the opacity of learned control policies, which are difficult to verify and understand. This has led regulators to require greater transparency in AI systems used in critical domains, e.g. the [48] and the [11] . For example, the EU AI Act specifies in Article 13.3 “The instructions for use shall contain [...] technical capabilities and characteristics of the high-risk AI system to provide information relevant to explain its output.” Explainable RL (XRL) methods [22] address this by  \nx  \n1 2 3 4 5  \nFigure 1: DT vs. Oblique DT.  \nproviding human-interpretable explanations of RL policies. Explanations are often symbolic models, such as decision trees [2], which can be verified when sufficiently small, paving the way for deploying  \nPreprint.  \nRL in safety-critical domains. However, much XRL work targets tasks with discrete control actions, as in early deep RL applications with discrete (yet very large) state spaces [24, 40] .  \nIn this paper, we address the challenge of learning small, interpretable surrogate models of RL policies with continuous action spaces and continuous-discrete state spaces. We propose ORCAID (Oblique Rule-Based Continuous-Action Interpretation for Deep RL Policies) to construct rulebased models from RL policies. ORCAID first partitions the state space by learning an oblique decision tree [27](extended for discrete/categorical features) . Unlike axis-aligned trees, oblique trees use hyperplanes that are linear combinations of features, allowing them to capture non-orthogonal boundaries compactly. As illustrated in Figure 1, an oblique decision tree can approximate a curvilinear bounda","cbCaiakh5xXF9QUA","https://ap.wps.com/l/cbCaiakh5xXF9QUA","pdf",863682,1,33,"English","en",105,"# Abstract\n# Introduction\n# Related Work","[{\"question\":\"What problem does ORCAID address in deep reinforcement learning?\",\"answer\":\"ORCAID targets the opacity of deep RL policies by producing interpretable rule-based surrogate models, especially for continuous action spaces and mixed continuous–discrete states.\"},{\"question\":\"How does ORCAID build an interpretable model from a trained deep RL policy?\",\"answer\":\"ORCAID first learns an oblique decision tree that partitions the state space using hyperplanes, then fits local linear models to represent each action dimension within tree leaves.\"},{\"question\":\"What techniques does ORCAID use to keep the rule set compact?\",\"answer\":\"It applies a three-stage split search (efficient random initialization, local refinement, backward elimination) and merges adjacent leaves to produce a concise set of interpretable rules.\"}]",1784199916,83,{"code":4,"msg":30,"data":31},"ok",{"site_id":24,"language":23,"slug":32,"title":13,"keywords":33,"description":14,"schema_data":34,"social_meta":86,"head_meta":88,"extra_data":90,"updated_unix":27},"orcaid-oblique-rule-based-continuous-action-interpretation-for-deep-rl-policies","",{"@graph":35,"@context":85},[36,53,68],{"@type":37,"itemListElement":38},"BreadcrumbList",[39,43,47,50],{"item":40,"name":41,"@type":42,"position":20},"https://docshare.wps.com","Home","ListItem",{"item":44,"name":45,"@type":42,"position":46},"https://docshare.wps.com/document/","Document",2,{"item":48,"name":12,"@type":42,"position":49},"https://docshare.wps.com/document/research-report/",3,{"item":51,"name":13,"@type":42,"position":52},"https://docshare.wps.com/document/orcaid-oblique-rule-based-continuous-action-interpretation-for-deep-rl-policies/84975/",4,{"url":51,"name":13,"@type":54,"author":55,"headline":13,"publisher":57,"fileFormat":60,"inLanguage":23,"description":14,"dateModified":61,"datePublished":62,"encodingFormat":60,"isAccessibleForFree":63,"interactionStatistic":64},"DigitalDocument",{"name":9,"@type":56},"Person",{"url":40,"name":58,"@type":59},"DocShare","Organization","application/pdf","2026-07-17","2026-07-16",true,{"@type":65,"interactionType":66,"userInteractionCount":20},"InteractionCounter",{"@type":67},"ViewAction",{"@type":69,"mainEntity":70},"FAQPage",[71,77,81],{"name":72,"@type":73,"acceptedAnswer":74},"What problem does ORCAID address in deep reinforcement learning?","Question",{"text":75,"@type":76},"ORCAID targets the opacity of deep RL policies by producing interpretable rule-based surrogate models, especially for continuous action spaces and mixed continuous–discrete states.","Answer",{"name":78,"@type":73,"acceptedAnswer":79},"How does ORCAID build an interpretable model from a trained deep RL policy?",{"text":80,"@type":76},"ORCAID first learns an oblique decision tree that partitions the state space using hyperplanes, then fits local linear models to represent each action dimension within tree leaves.",{"name":82,"@type":73,"acceptedAnswer":83},"What techniques does ORCAID use to keep the rule set compact?",{"text":84,"@type":76},"It applies a three-stage split search (efficient random initialization, local refinement, backward elimination) and merges adjacent leaves to produce a concise set of interpretable rules.","https://schema.org",{"og:url":51,"og:type":87,"og:title":13,"og:site_name":58,"og:description":14},"article",{"robots":89,"canonical":51},"index,follow",{"doc_id":7,"site_id":24},{"code":4,"msg":5,"data":92},[93,97,101,105,110,115,120,123,128,131,135],{"id":20,"doc_module":4,"doc_module_name":45,"category_name":94,"show_sort_weight":95,"slug":96},"Story & Novel",90,"story-novel",{"id":46,"doc_module":4,"doc_module_name":45,"category_name":98,"show_sort_weight":99,"slug":100},"Literature",80,"literature",{"id":52,"doc_module":4,"doc_module_name":45,"category_name":102,"show_sort_weight":103,"slug":104},"Exam",70,"exam",{"id":106,"doc_module":4,"doc_module_name":45,"category_name":107,"show_sort_weight":108,"slug":109},5,"Comic",60,"comic",{"id":111,"doc_module":4,"doc_module_name":45,"category_name":112,"show_sort_weight":113,"slug":114},6,"Technology",50,"technology",{"id":116,"doc_module":4,"doc_module_name":45,"category_name":117,"show_sort_weight":118,"slug":119},7,"Healthcare",40,"healthcare",{"id":11,"doc_module":4,"doc_module_name":45,"category_name":12,"show_sort_weight":121,"slug":122},30,"research-report",{"id":124,"doc_module":4,"doc_module_name":45,"category_name":125,"show_sort_weight":126,"slug":127},9,"Religion & Spirituality",20,"religion-spirituality",{"id":126,"doc_module":4,"doc_module_name":45,"category_name":129,"show_sort_weight":126,"slug":130},"World Cup","world-cup",{"id":132,"doc_module":4,"doc_module_name":45,"category_name":133,"show_sort_weight":132,"slug":134},10,"Lifestyle","lifestyle",{"id":136,"doc_module":4,"doc_module_name":45,"category_name":137,"show_sort_weight":106,"slug":138},19,"General","general"]