{"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/disentangling-sources-of-risk-for","title":"Disentangling Sources of Risk for Distributional Multi-Agent Reinforcement Learning","arxiv_id":null,"date":"2021-09-29","proceeding":null,"authors":["Kyunghwan Son","Junsu Kim","Yung Yi","Jinwoo Shin"],"abstract":"In cooperative multi-agent reinforcement learning, state transitions, rewards, and actions can all induce randomness (or uncertainty) in the observed long-term returns. These randomnesses are reflected from two risk sources: (a) agent-wise risk (i.e., how cooperative our teammates act for a given agent) and (b) environment-wise risk (i.e., transition stochasticity). Although these two sources are both important factors for learning robust policies of agents, prior works do not separate them or deal with only a single risk source, which could lead to suboptimal equilibria. In this paper, we propose Disentangled RIsk-sensitive Multi-Agent reinforcement learning (DRIMA), a novel framework being capable of disentangling risk sources. Our main idea is to separate risk-level leverages (i.e., quantiles) in both centralized training and decentralized execution with a hierarchical quantile structure and quantile regression. Our experiments demonstrate that DRIMA significantly outperforms prior-arts across various scenarios in StarCraft Multi-agent Challenge. Notably, DRIMA shows robust performance regardless of reward shaping, exploration schedule, where prior methods learn only a suboptimal policy.","url_abs":"https://openreview.net/forum?id=5qwA7LLbgP0","url_pdf":"https://openreview.net/pdf?id=5qwA7LLbgP0","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":[],"tasks":[{"task_slug":"multi-agent-reinforcement-learning","task_name":"Multi-agent Reinforcement Learning"},{"task_slug":"reinforcement-learning","task_name":"Reinforcement Learning"},{"task_slug":"reinforcement-learning-1","task_name":"Reinforcement Learning (RL)"},{"task_slug":"smac-1","task_name":"SMAC+"},{"task_slug":"starcraft","task_name":"Starcraft"},{"task_slug":"quantile-regression","task_name":"quantile regression"},{"task_slug":"reinforcement-learning-2","task_name":"reinforcement-learning"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[{"leaderboard":"/sota/smac-on-smac-def-armored-parallel","task":"SMAC+","dataset":"Def_Armored_parallel","model":"DRIMA","rank_in_archive_order":3,"of":10,"metrics":{"Median Win Rate":"60.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-def-armored-sequential","task":"SMAC+","dataset":"Def_Armored_sequential","model":"DRIMA","rank_in_archive_order":1,"of":11,"metrics":{"Median Win Rate":"100"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-def-infantry-parallel","task":"SMAC+","dataset":"Def_Infantry_parallel","model":"DRIMA","rank_in_archive_order":2,"of":10,"metrics":{"Median Win Rate":"100.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-def-infantry-sequential","task":"SMAC+","dataset":"Def_Infantry_sequential","model":"DRIMA","rank_in_archive_order":2,"of":11,"metrics":{"Median Win Rate":"100"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-def-outnumbered-parallel","task":"SMAC+","dataset":"Def_Outnumbered_parallel","model":"DRIMA","rank_in_archive_order":1,"of":10,"metrics":{"Median Win Rate":"70.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-def-outnumbered-sequential","task":"SMAC+","dataset":"Def_Outnumbered_sequential","model":"DRIMA","rank_in_archive_order":1,"of":11,"metrics":{"Median Win Rate":"100"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-complicated-parallel","task":"SMAC+","dataset":"Off_Complicated_parallel","model":"DRIMA","rank_in_archive_order":1,"of":10,"metrics":{"Median Win Rate":"100"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-complicated-sequential","task":"SMAC+","dataset":"Off_Complicated_sequential","model":"DRIMA","rank_in_archive_order":1,"of":4,"metrics":{"Median Win Rate":"96.9"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-distant-parallel","task":"SMAC+","dataset":"Off_Distant_parallel","model":"DRIMA","rank_in_archive_order":1,"of":10,"metrics":{"Median Win Rate":"95.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-distant-sequential","task":"SMAC+","dataset":"Off_Distant_sequential","model":"DRIMA","rank_in_archive_order":1,"of":4,"metrics":{"Median Win Rate":"100"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-hard-parallel","task":"SMAC+","dataset":"Off_Hard_parallel","model":"DRIMA","rank_in_archive_order":1,"of":10,"metrics":{"Median Win Rate":"80.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-hard-sequential","task":"SMAC+","dataset":"Off_Hard_sequential","model":"DRIMA","rank_in_archive_order":2,"of":4,"metrics":{"Median Win Rate":"93.8"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-near-parallel","task":"SMAC+","dataset":"Off_Near_parallel","model":"DRIMA","rank_in_archive_order":2,"of":10,"metrics":{"Median Win Rate":"95.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-near-sequential","task":"SMAC+","dataset":"Off_Near_sequential","model":"DRIMA","rank_in_archive_order":1,"of":4,"metrics":{"Median Win Rate":"93.8"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-superhard-parallel","task":"SMAC+","dataset":"Off_Superhard_parallel","model":"DRIMA","rank_in_archive_order":10,"of":10,"metrics":{"Median Win Rate":"0.0"},"uses_additional_data":false},{"leaderboard":"/sota/smac-on-smac-off-superhard-sequential","task":"SMAC+","dataset":"Off_Superhard_sequential","model":"DRIMA","rank_in_archive_order":1,"of":4,"metrics":{"Median Win Rate":"15.6"},"uses_additional_data":false}],"syntology":{"syntology_url":null,"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}