[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"detail-sidebar-cat-0-en-105":3,"doc-seo-315098-105":59,"doc-detail-315098-en":131},{"code":4,"msg":5,"data":6},0,"success",[7,13,18,23,28,33,38,43,48,51,55],{"id":8,"doc_module":4,"doc_module_name":9,"category_name":10,"show_sort_weight":11,"slug":12},1,"Document","Story & Novel",90,"story-novel",{"id":14,"doc_module":4,"doc_module_name":9,"category_name":15,"show_sort_weight":16,"slug":17},2,"Literature",80,"literature",{"id":19,"doc_module":4,"doc_module_name":9,"category_name":20,"show_sort_weight":21,"slug":22},4,"Exam",70,"exam",{"id":24,"doc_module":4,"doc_module_name":9,"category_name":25,"show_sort_weight":26,"slug":27},5,"Comic",60,"comic",{"id":29,"doc_module":4,"doc_module_name":9,"category_name":30,"show_sort_weight":31,"slug":32},6,"Technology",50,"technology",{"id":34,"doc_module":4,"doc_module_name":9,"category_name":35,"show_sort_weight":36,"slug":37},7,"Healthcare",40,"healthcare",{"id":39,"doc_module":4,"doc_module_name":9,"category_name":40,"show_sort_weight":41,"slug":42},8,"Research & Report",30,"research-report",{"id":44,"doc_module":4,"doc_module_name":9,"category_name":45,"show_sort_weight":46,"slug":47},9,"Religion & Spirituality",20,"religion-spirituality",{"id":46,"doc_module":4,"doc_module_name":9,"category_name":49,"show_sort_weight":46,"slug":50},"World Cup","world-cup",{"id":52,"doc_module":4,"doc_module_name":9,"category_name":53,"show_sort_weight":52,"slug":54},10,"Lifestyle","lifestyle",{"id":56,"doc_module":4,"doc_module_name":9,"category_name":57,"show_sort_weight":24,"slug":58},19,"General","general",{"code":4,"msg":60,"data":61},"ok",{"site_id":62,"language":63,"slug":64,"title":65,"keywords":66,"description":67,"schema_data":68,"social_meta":124,"head_meta":126,"extra_data":128,"updated_unix":130},105,"en","control-of-memory-active-perception-and-action-in-minecraft","Control of Memory, Active Perception, and Action in Minecraft","","Reinforcement learning in Minecraft is used to introduce controllable benchmarks targeting partial observability from first-person vision, delayed rewards, high-dimensional visual inputs, and the requirement for correct active perception. The paper compares existing deep reinforcement learning architectures with new memory-based architectures tailored to these cognitive-like demands. It additionally tests generalization on environments and Minecraft maps not encountered during training. Results show improved generalization to unseen settings for the proposed memory mechanisms.",{"@graph":69,"@context":123},[70,84,106],{"@type":71,"itemListElement":72},"BreadcrumbList",[73,77,79,82],{"item":74,"name":75,"@type":76,"position":8},"https://docshare.wps.com","Home","ListItem",{"item":78,"name":9,"@type":76,"position":14},"https://docshare.wps.com/document/",{"item":80,"name":40,"@type":76,"position":81},"https://docshare.wps.com/document/research-report/",3,{"item":83,"name":65,"@type":76,"position":19},"https://docshare.wps.com/document/control-of-memory-active-perception-and-action-in-minecraft/315098/",{"url":83,"name":65,"@type":85,"image":86,"author":91,"headline":65,"publisher":94,"fileFormat":97,"inLanguage":63,"description":67,"dateModified":98,"datePublished":99,"encodingFormat":97,"isAccessibleForFree":100,"interactionStatistic":101},"DigitalDocument",{"url":87,"@type":88,"width":89,"height":90},"https://docshare.wps.com/thumbnails/control-of-memory-active-perception-and-action-in-minecraft/315098.png","ImageObject",300,407,{"name":92,"@type":93},"Rizky","Person",{"url":74,"name":95,"@type":96},"DocShare","Organization","application/pdf","2026-11-07","2026-09-20",true,{"@type":102,"interactionType":103,"userInteractionCount":105},"InteractionCounter",{"@type":104},"ViewAction",16,{"@type":107,"mainEntity":108},"FAQPage",[109,115,119],{"name":110,"@type":111,"acceptedAnswer":112},"What reinforcement learning challenges does the paper’s Minecraft tasks emphasize?","Question",{"text":113,"@type":114},"The tasks emphasize partial observability from first-person visual observations, delayed rewards, high-dimensional visual observations, and the need to use active perception correctly to succeed.","Answer",{"name":116,"@type":111,"acceptedAnswer":117},"How do the proposed memory-based DRL architectures differ from existing ones?",{"text":118,"@type":114},"The proposed architectures store recent observations in memory and retrieve relevant information conditioned on temporal context, unlike existing architectures whose memory retrieval is not context-conditioned.",{"name":120,"@type":111,"acceptedAnswer":121},"Do the proposed architectures generalize better than existing DRL methods?",{"text":122,"@type":114},"Yes. Experimental results indicate that the proposed architectures generalize to unseen environments and larger/unseen Minecraft maps better than existing deep reinforcement learning architectures.","https://schema.org",{"og:url":83,"og:type":125,"og:title":65,"og:site_name":95,"og:description":67},"article",{"robots":127,"canonical":83},"index,follow",{"doc_id":129,"site_id":62},315098,1789964442,{"code":4,"msg":5,"data":132},{"doc_id":129,"user_id":133,"nickname":92,"user_avatar":134,"doc_module":4,"category_id":39,"category_name":40,"doc_title":65,"doc_description":67,"doc_content":135,"file_id":136,"file_url":137,"file_type":138,"file_size":139,"view_count":105,"is_deleted":4,"is_public":8,"is_downloadable":8,"audit_status":8,"page_count":52,"language":140,"language_code":63,"site_id":62,"html_lang":63,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":67,"update_tm":144,"read_time":145,"preview_page_count":4},962085564807,"https://ap-avatar.wpscdn.com/davatar_6f874abed73319feea01a86fa6f0fab8","Control of Memory, Active Perception, and Action in Minecraft  \nJunhyuk Oh JUNHYUK @UMICH . EDU  \nValliappa Chockalingam VALLI @UMICH . EDU  \nSatinder Singh BAVEJA @UMICH . EDU  \nHonglak Lee HONGLAK @UMICH . EDU  \nComputer Science & Engineering, University of Michigan  \nAbstract  \nIn this paper, we introduce a new set of reinforcement learning (RL) tasks in Minecraft (a flexible 3D world) . We then use these tasks to systematically compare and contrast existing deep reinforcement learning (DRL) architectures with our new memory-based DRL architectures. These tasks are designed to emphasize, in a controllable manner, issues that pose challenges for RL methods including partial observability (due to first-person visual observations), delayed rewards, high-dimensional visual observations, and the need to use active perception ina correct manner so as to perform well in the tasks. While these tasks are conceptually simple to describe, by virtue of having all of these challenges simultaneously they are difficult for current DRL architectures. Additionally, we evaluate the generalization performance of the architectures on environments not used during training. The experimental results show that our new architectures generalize to unseen environments better than existing DRL architectures.  \n1. Introduction  \nDeep learning approaches (surveyed in LeCun et al., 2015 ; Schmidhuber, 2015) have made advances in many lowlevel perceptual supervised learning problems (Krizhevsky et al., 2012 ; Girshick et al., 2014 ; Simonyan & Zisserman, 2015) . This success has been extended to reinforcement learning (RL) problems that involve visual perception. For example, the Deep Q-Network (DQN) (Mnih et al., 2015) architecture has been shown to successfully learn to play many Atari 2600 games in the Arcade Learning Environment (ALE) benchmark (Bellemare et al., 2013) by learning visual features useful for control directly from raw pixels using Q-Learning (Watkins & Dayan, 1992) .  \nProceedings of the 33 rd International Conference on Machine Learning, New York, NY, USA, 2016 . JMLR: W&CP volume 48. Copyright 2016 by the author(s) .  \n(a) t=3 (b) t=10 (c) t=11 (d) t=19 Figure 1 . Example task in Minecraft. In this task, the agent should visit the red block if the indicator (next to the start location) is yellow. Otherwise, if the indicator is green, it should visit the blue block. The top row shows the agent's first-person observation. The bottom row visualizes the map and the agent's location; this is not available to the agent. (a) The agent observes the yellow indicator. (b) The agent looks left and sees the blue block, (c) but it decides to keep going straight having previously seen the yellow indicator. (d) Finally, it visits the red block and receives a positive reward.  \nRecently, researchers have explored problems that require faculties associated with higher-level cognition (e.g., inferring simple general purpose algorithms: Graves et al., 2014, and, Q&A: Weston et al., 2015) . Most of these advances, however, are restricted to the supervised learning setting, which provides clear error signals. In this paper, we are interested in extending this success to similarly cognition-inspired RL tasks. Specifically, this paper introduces a set of tasks in Minecraft 1 , a flexible 3D world in which an agent can collect resources, build structures, and survive attacks from enemies. Our RL tasks (one example is illustrated in Figure 1) not only have the usual RL challenges of partial observability, high-dimensional (visual) perception, and delayed reward, but also require an agent to develop movement policies by learning how to use its active perception to observe useful information and collect reward. In addition, our RL tasks require an agent to learn to use any memory it possesses including its interaction with active perception which feeds observations into  \n1 [https://minecraft.net/](https://minecraft.net/)  \nmemory. We note that for simplicity we here","cbCaicShBsJ0FrNg","https://ap.wps.com/l/cbCaicShBsJ0FrNg","pdf",1330148,"English","# Introduction\n# Related Work","[{\"question\":\"What reinforcement learning challenges does the paper’s Minecraft tasks emphasize?\",\"answer\":\"The tasks emphasize partial observability from first-person visual observations, delayed rewards, high-dimensional visual observations, and the need to use active perception correctly to succeed.\"},{\"question\":\"How do the proposed memory-based DRL architectures differ from existing ones?\",\"answer\":\"The proposed architectures store recent observations in memory and retrieve relevant information conditioned on temporal context, unlike existing architectures whose memory retrieval is not context-conditioned.\"},{\"question\":\"Do the proposed architectures generalize better than existing DRL methods?\",\"answer\":\"Yes. Experimental results indicate that the proposed architectures generalize to unseen environments and larger/unseen Minecraft maps better than existing deep reinforcement learning architectures.\"}]","Control of Memory, Active Perception, and Action in Minecraft | PDF",1789903251,25]