Update README.md

README.md

@@ -10,11 +10,12 @@ library_name: pytorch
 license: gpl-3.0
 ---
 
-# M<sup>3</sup>: A Modular World Model over Streams of Tokens
+# Simulus: Combining Improvements in Sample-Efficient World Model Agents
 
-📄 [Paper](https://arxiv.org/abs/2502.11537) ▪️ 💾 [Code](https://github.com/leor-c/M3) ▪️ 🧠 [Trained Model Weights](https://huggingface.co/leorc/M3)
+📄 [Paper](https://arxiv.org/abs/2502.11537) ▪️ 💾 [Code](https://github.com/leor-c/Simulus) ▪️ 🧠 [Trained Model Weights](https://huggingface.co/leorc/Simulus)
 
-M<sup>3</sup> is a modular world model that extends the token-based world model framework to handle diverse observation and action modalities through independent, modality-specific components.  It incorporates improvements from existing literature to enhance agent performance and achieves state-of-the-art sample efficiency for planning-free world models.  It is the first method of this kind to reach a human-level median score on Atari 100K, exhibiting superhuman performance on 13 games.  The model weights provided here cover Atari 100K, DeepMind Control Suite Proprioceptive 500K, and Craftax (Symbolic) 1M.
+
+World models (WMs) represent the frontier of sample-efficient reinforcement learning, but their complexity leaves many promising improvements unrealized due to the significant expertise and effort required to identify and integrate them. Inspired by Rainbow, which showed that individually known improvements to DQN complement each other and can be effectively combined, we take on this challenge and ask whether the same principle applies to world model agents. We introduce Simulus, a modular token-based WM agent that integrates: (1) a flexible tokenization framework supporting arbitrary combinations of observation and action modalities; (2) intrinsic motivation for epistemic uncertainty reduction; (3) prioritized world model replay; and (4) regression-as-classification for reward and return prediction. Simulus achieves state-of-the-art sample efficiency for planning-free WMs across three diverse benchmarks: visual Atari 100K, continuous-control DMC Proprioception 500K, and symbolic Craftax-1M. Notably, intrinsic motivation proves beneficial even under the tight interaction budgets of sample-efficient RL, despite the risk of wasting scarce interactions on task-irrelevant experience. Ablation studies reveal that each component contributes individually, and their combination yields synergistic gains.
 
 <div align="center">
 <img src="https://github.com/user-attachments/assets/14734453-38dd-4bc0-a2e0-349e4eec37a2" height="220" />