{"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/what-makes-a-good-diffusion-planner-for","title":"What Makes a Good Diffusion Planner for Decision Making?","arxiv_id":null,"date":"2025-03-01","proceeding":"ICLR 2025 3","authors":["Haofei Lu","Dongqi Han","Yifei Shen","Dongsheng Li"],"abstract":"Diffusion models have recently shown significant potential in solving\r\ndecision-making problems, particularly in generating behavior plans -- also\r\nknown as diffusion planning. While numerous studies have demonstrated the\r\nimpressive performance of diffusion planning, the mechanisms behind the key\r\ncomponents of a good diffusion planner remain unclear and the design choices\r\nare highly inconsistent in existing studies. In this work, we address this\r\nissue through systematic empirical experiments on diffusion planning in an\r\noffline reinforcement learning (RL) setting, providing practical insights into\r\nthe essential components of diffusion planning. We trained and evaluated over\r\n6,000 diffusion models, identifying the critical components such as guided\r\nsampling, network architecture, action generation and planning strategy. We\r\nrevealed that some design choices opposite to the common practice in previous\r\nwork in diffusion planning actually lead to better performance, e.g.,\r\nunconditional sampling with selection can be better than guided sampling and\r\nTransformer outperforms U-Net as denoising network. Based on these insights, we\r\nsuggest a simple yet strong diffusion planning baseline that achieves\r\nstate-of-the-art results on standard offline RL benchmarks.","url_abs":"https://openreview.net/forum?id=7BQkXXM8Fy","url_pdf":"https://openreview.net/pdf?id=7BQkXXM8Fy","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":"what-makes-a-good-diffusion-planner-for","repo_url":"https://github.com/Josh00-Lu/DiffusionVeteran","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":0,"framework":"pytorch","reach":{"status":"ok","spdx":"Apache-2.0"}}],"tasks":[{"task_slug":"action-generation","task_name":"Action Generation"},{"task_slug":"decision-making","task_name":"Decision Making"},{"task_slug":"denoising","task_name":"Denoising"},{"task_slug":"offline-rl","task_name":"Offline RL"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"}],"methods":[{"method_slug":"concatenated-skip-connection","method_name":"Concatenated Skip Connection"},{"method_slug":"convolution","method_name":"Convolution"},{"method_slug":"diffusion","method_name":"Diffusion"},{"method_slug":"max-pooling","method_name":"Max Pooling"},{"method_slug":"relu","method_name":"ReLU"},{"method_slug":"u-net","method_name":"U-Net"}],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}