{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/training-frankensteins-creature-to-stack","title":"The CoSTAR Block Stacking Dataset: Learning with Workspace Constraints","arxiv_id":"1810.11714","date":"2018-10-27","proceeding":null,"authors":["Andrew Hundt","Varun Jain","Chia-Hung Lin","Chris Paxton","Gregory D. Hager"],"abstract":"A robot can now grasp an object more effectively than ever before, but once\nit has the object what happens next? We show that a mild relaxation of the task\nand workspace constraints implicit in existing object grasping datasets can\ncause neural network based grasping algorithms to fail on even a simple block\nstacking task when executed under more realistic circumstances.\n  To address this, we introduce the JHU CoSTAR Block Stacking Dataset (BSD),\nwhere a robot interacts with 5.1 cm colored blocks to complete an\norder-fulfillment style block stacking task. It contains dynamic scenes and\nreal time-series data in a less constrained environment than comparable\ndatasets. There are nearly 12,000 stacking attempts and over 2 million frames\nof real data. We discuss the ways in which this dataset provides a valuable\nresource for a broad range of other topics of investigation.\n  We find that hand-designed neural networks that work on prior datasets do not\ngeneralize to this task. Thus, to establish a baseline for this dataset, we\ndemonstrate an automated search of neural network based models using a novel\nmultiple-input HyperTree MetaModel, and find a final model which makes\nreasonable 3D pose predictions for grasping and stacking on our dataset.\n  The CoSTAR BSD, code, and instructions are available at\nhttps://sites.google.com/site/costardataset.","url_abs":"http://arxiv.org/abs/1810.11714v2","url_pdf":"http://arxiv.org/pdf/1810.11714v2.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"training-frankensteins-creature-to-stack","repo_url":"https://github.com/ahundt/enas","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"training-frankensteins-creature-to-stack","repo_url":"https://github.com/jhu-lcsr/costar_plan","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":1,"framework":"tf","reach":null},{"paper_slug":"training-frankensteins-creature-to-stack","repo_url":"https://github.com/ahundt/costar_dataset","is_official":1,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":null}],"tasks":[{"task_slug":"6d-pose-estimation-using-rgbd","task_name":"6D Pose Estimation using RGBD"},{"task_slug":"industrial-robots","task_name":"Industrial Robots"},{"task_slug":"architecture-search","task_name":"Neural Architecture Search"},{"task_slug":"object","task_name":"Object"},{"task_slug":"offline-rl","task_name":"Offline RL"},{"task_slug":"robot-task-planning","task_name":"Robot Task Planning"},{"task_slug":"robotic-grasping","task_name":"Robotic Grasping"},{"task_slug":"time-series","task_name":"Time Series Analysis"}],"methods":[{"method_slug":"hypertree-metamodel","method_name":"HyperTree MetaModel"}],"datasets_introduced":[{"slug":"jhu-costar-block-stacking-dataset","name":"JHU CoSTAR Block Stacking Dataset","full_name":""}],"methods_introduced":[{"slug":"hypertree-metamodel","name":"HyperTree MetaModel","full_name":"HyperTree MetaModel"}],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=1810.11714","mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}