AI Paper Atlas

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

Dreaming in Code for Curriculum Learning in Open-Ended Worlds

Konstantinos Mitsides, Maxence Faldor, Antoine Cully

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
add FM-generated code-level curriculum (DiCode) to PPO-GTrXLPPO-GTrXL on Craftax · Craftax held-out testfinal mean return: 41.59 -> 48.55Improved
add learnability-prioritized parent selection instead of random parent sampling (ablation)DiCode · Craftax held-out testfinal mean return: 46.44 -> 48.55Improved
add learnability-prioritized parent selection (late-game combat) (ablation)DiCode · Craftax Gnome Warrior / Archer successsuccess rate: Gnome Warrior: 0.6% -> 14.7%, Gnome Archer: 0.1% -> 11.2%Improved
add environment reshaping (initial states and dynamics) beyond goal selection (ablation)DiCode · Craftax held-out testfinal mean return: 42.82 -> 48.55Improved
use larger generator FM (Qwen 80B instead of 30B) (ablation)DiCode · Craftax held-out testfinal mean return: 44.38 -> 45.88Improved
use Qwen3-235B instead of Qwen 80B as generator (ablation)DiCode · Craftax held-out testfinal mean return: 45.88 -> 48.55Improved
add closed-loop grounding (parent description and performance profiles in FM context) (ablation)DiCode · Craftax held-out testfinal mean return: 41.12 -> 48.55Improved
open-loop generation versus PPO-GTrXL baseline (ablation)DiCode-OL · Craftax held-out testfinal mean return: 41.59 -> 41.12No clear effect
DiCode versus best baseline on Make Iron ArmourPPO-GTrXL baselines · Craftax Make Iron Armoursuccess rate: 15% -> 47%Improved
DiCode versus strongest baseline on entering Gnomish MinesPPO-GTrXL baselines · Craftax Floor 2 entryfraction of episodes: 9% -> 32%Improved
DiCode versus baselines on late-game combatPPO-GTrXL baselines · Craftax Gnome Warrior / Archersuccess rate: 0% -> 15% (Gnome Warrior), 0% -> 11% (Gnome Archer)Improved
DiCode versus baseline on Make Diamond SwordPPO-GTrXL baselines · Craftax Make Diamond Swordsuccess rate: 3% -> 6%Improved

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