AI Paper Atlas

Research map / Multi-agent systems and games / Human-AI and goal-conditioned

Scaling Multi-Agent Environment Co-Design with Diffusion Models

Hao Xiang Li, Michael Amir, Amanda Prorok

ICML 2026 · Multi-agent systems and games › Human-AI and goal-conditioned · approach: World-model-based agents · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
co-design the environment with DiCoDe versus a fixed environmentMAPPO agents · D-RWARE Cornerepisode reward: 9.6 -> 12.1 (26%)Improved
use DiCoDe instead of RL co-design baselineco-design · D-RWARE Cornerboxes delivered: 8.7 -> 12.1; 39% more with 66% fewer samplesImproved
replace Projected Universal Guidance with gradient descent (ablation)DiCoDe · D-RWARE Cornerepisode reward: 12.1 -> 9.3Hurt
replace PUG with top-k sampling (ablation)DiCoDe · D-RWARE Cornerepisode reward: 12.1 -> 8.2Hurt
replace PUG with the ADD diffusion guidance (ablation)DiCoDe · D-RWARE Cornerepisode reward: 12.1 -> 9.1Hurt
add critic distillation instead of training environment critic on trajectory returns (ablation)DiCoDe · D-RWARE Cornerepisode reward: 10.9 -> 12.1Improved
use shelf coordinate representation instead of binary mask (ablation)DiCoDe · D-RWARE Cornerepisode reward: 12.1 -> 11.7No clear effect
add constraint projection to universal guidance (PUG vs UG) (ablation)environment search with fixed critic · D-RWARE Cornercritic-estimated valueNo clear effect
use learned guided diffusion rather than best of 1024 uniformly sampled environments (ablation)environment design · D-RWARE Cornercritic-estimated value: best-of-1024 sampling is 12% worse than mean of PUGImproved
use DiCoDe instead of RL co-design as turbines increaseco-design · WFCRL 2 to 16 turbinesepisode reward: WFCRL16 254 -> 282Improved
use DiCoDe on WFCRL and ONav versus baselinesco-design · WFCRL2/4/8/16 and ONavepisode reward: 9.5% above Fixed, 10.3% above RL, 17.1% above DRImproved
use DiCoDe versus gradient descent ablation on ONav (ablation)DiCoDe · VMAS ONavepisode reward: 2.29 vs 2.29No clear effect

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy