# HarmonyDream: Task Harmonization Inside World Models

arXiv: https://arxiv.org/abs/2310.00344 · HTML: https://mlsanity.com/paper/2310.00344/

## Key facts

- Cited by 22 papers in the map (data through 14 Oct 2026).
- Placed in Latent dynamics models and model-based RL, one of 11 directions in World models.
- Builds on 53 of its 60 references that are in the map.
- First posted on arXiv 30 Sep 2023.

## Placed in

- World models / [Latent dynamics models and model-based RL](https://mlsanity.com/wm/latent-dynamics-models-and-model-based-rl/)
- LLM alignment / RLHF / preference optimisation / [Reward modeling and RLHF](https://mlsanity.com/alignment/reward-modeling-and-rlhf/)

## Builds on (most cited first, 30 of 53)

- [PyTorch: An Imperative Style, High-Performance Deep Learning Library](https://mlsanity.com/paper/1912.01703/) (2019)
- [Unsupervised Learning of Visual Features by Contrasting Cluster Assignments](https://mlsanity.com/paper/2006.09882/) (2020)
- [Shortcut Learning in Deep Neural Networks](https://mlsanity.com/paper/2004.07780/) (2020)
- [Offline Reinforcement Learning: Tutorial, Review, and Perspectives on Open Problems](https://mlsanity.com/paper/2005.01643/) (2020)
- [Dream to Control: Learning Behaviors by Latent Imagination](https://mlsanity.com/paper/1912.01603/) (2019)
- [Progressive Neural Networks](https://mlsanity.com/paper/1606.04671/) (2016)
- [Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics](https://mlsanity.com/paper/1705.07115/) (2017)
- [Mastering Atari, Go, Chess and Shogi by Planning with a Learned Model](https://mlsanity.com/paper/1911.08265/) (2019)
- [Learning Latent Dynamics for Planning from Pixels](https://mlsanity.com/paper/1811.04551/) (2018)
- [Gradient Surgery for Multi-Task Learning](https://mlsanity.com/paper/2001.06782/) (2020)
- [Mastering Diverse Domains through World Models](https://mlsanity.com/paper/2301.04104/) (2023)
- [Meta-World: A Benchmark and Evaluation for Multi-Task and Meta Reinforcement Learning](https://mlsanity.com/paper/1910.10897/) (2019)
- [Mixed Precision Training](https://mlsanity.com/paper/1710.03740/) (2017)
- [Mastering Atari with Discrete World Models](https://mlsanity.com/paper/2010.02193/) (2020)
- [DeepMind Control Suite](https://mlsanity.com/paper/1801.00690/) (2018)
- [GradNorm: Gradient Normalization for Adaptive Loss Balancing in Deep Multitask Networks](https://mlsanity.com/paper/1711.02257/) (2017)
- [An Overview of Multi-Task Learning in Deep Neural Networks](https://mlsanity.com/paper/1706.05098/) (2017)
- [Recurrent World Models Facilitate Policy Evolution](https://mlsanity.com/paper/1809.01999/) (2018)
- [CURL: Contrastive Unsupervised Representations for Reinforcement Learning](https://mlsanity.com/paper/2004.04136/) (2020)
- [RLBench: The Robot Learning Benchmark & Learning Environment](https://mlsanity.com/paper/1909.12271/) (2019)
- [End-to-End Multi-Task Learning with Attention](https://mlsanity.com/paper/1803.10704/) (2018)
- [Model-Based Reinforcement Learning for Atari](https://mlsanity.com/paper/1903.00374/) (2019)
- [MineDojo: Building Open-Ended Embodied Agents with Internet-Scale Knowledge](https://mlsanity.com/paper/2206.08853/) (2022)
- [Cross-stitch Networks for Multi-task Learning](https://mlsanity.com/paper/1604.03539/) (2016)
- [Conflict-Averse Gradient Descent for Multi-task Learning](https://mlsanity.com/paper/2110.14048/) (2021)
- [Reinforcement Learning with Unsupervised Auxiliary Tasks](https://mlsanity.com/paper/1611.05397/) (2016)
- [DayDreamer: World Models for Physical Robot Learning](https://mlsanity.com/paper/2206.14176/) (2022)
- [Temporal Difference Learning for Model Predictive Control](https://mlsanity.com/paper/2203.04955/) (2022)
- [Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning](https://mlsanity.com/paper/2107.09645/) (2021)
- [Transformers are Sample-Efficient World Models](https://mlsanity.com/paper/2209.00588/) (2022)

## Built on by (most cited first, 22 of 22)

- [iVideoGPT: Interactive VideoGPTs are Scalable World Models](https://mlsanity.com/paper/2405.15223/) (2024)
- [Is Sora a World Simulator? A Comprehensive Survey on General World Models and Beyond](https://mlsanity.com/paper/2405.03520/) (2024)
- [A Survey: Learning Embodied Intelligence from Physical Simulators and World Models](https://mlsanity.com/paper/2507.00917/) (2025)
- [World Action Models: The Next Frontier in Embodied AI](https://mlsanity.com/paper/2605.12090/) (2026)
- [VideoWorld 2: Learning Transferable Knowledge from Real-world Videos](https://mlsanity.com/paper/2602.10102/) (2026)
- [Four Principles for Physically Interpretable World Models](https://mlsanity.com/paper/2503.02143/) (2025)
- [Discrete Codebook World Models for Continuous Control](https://mlsanity.com/paper/2503.00653/) (2025)
- [FOUNDER: Grounding Foundation Models in World Models for Open-Ended Embodied Decision Making](https://mlsanity.com/paper/2507.12496/) (2025)
- [CompilerDream: Learning a Compiler World Model for General Code Optimization](https://mlsanity.com/paper/2404.16077/) (2024)
- [Disentangled World Models: Learning to Transfer Semantic Knowledge from Distracting Videos for Reinforcement Learning](https://mlsanity.com/paper/2503.08751/) (2025)
- [SAMPO:Scale-wise Autoregression with Motion PrOmpt for generative world models](https://mlsanity.com/paper/2509.15536/) (2025)
- [Morphology-Conditioned World Model for Cross-Embodiment Quadrupedal Locomotion](https://mlsanity.com/paper/2604.08780/) (2026)
- [Representing Positional Information in Generative World Models for Object Manipulation](https://mlsanity.com/paper/2409.12005/) (2024)
- [EDELINE: Enhancing Memory in Diffusion-based World Models via Linear-Time Sequence Modeling](https://mlsanity.com/paper/2502.00466/) (2025)
- [One Model for All Tasks: Leveraging Efficient World Models in Multi-Task Planning](https://mlsanity.com/paper/2509.07945/) (2025)
- [OpenWorldLib: A Unified Codebase and Definition of Advanced World Models](https://mlsanity.com/paper/2604.04707/) (2026)
- [Efficient Reinforcement Learning Through Adaptively Pretrained Visual Encoder](https://mlsanity.com/paper/2502.05555/) (2025)
- [DyMoDreamer: World Modeling with Dynamic Modulation](https://mlsanity.com/paper/2509.24804/) (2025)
- [Mixture-of-World Models: Scaling Multi-Task Reinforcement Learning with Modular Latent Dynamics](https://mlsanity.com/paper/2602.01270/) (2026)
- [Pessimistic Auxiliary Policy for Offline Reinforcement Learning](https://mlsanity.com/paper/2602.23974/) (2026)
- [Probing the Impact of Scale on Data-Efficient, Generalist Transformer World Models for Atari](https://mlsanity.com/paper/2605.08578/) (2026)
- [Transferable Delay-Aware Reinforcement Learning via Implicit Causal Graph Modeling](https://mlsanity.com/paper/2605.12312/) (2026)

## Abstract (arXiv)

Model-based reinforcement learning (MBRL) holds the promise of sample-efficient learning by utilizing a world model, which models how the environment works and typically encompasses components for two tasks: observation modeling and reward modeling. In this paper, through a dedicated empirical investigation, we gain a deeper understanding of the role each task plays in world models and uncover the overlooked potential of sample-efficient MBRL by mitigating the domination of either observation or reward modeling. Our key insight is that while prevalent approaches of explicit MBRL attempt to restore abundant details of the environment via observation models, it is difficult due to the environment's complexity and limited model capacity. On the other hand, reward models, while dominating implicit MBRL and adept at learning compact task-centric dynamics, are inadequate for sample-efficient learning without richer learning signals. Motivated by these insights and discoveries, we propose a simple yet effective approach, HarmonyDream, which automatically adjusts loss coefficients to maintain task harmonization, i.e. a dynamic equilibrium between the two tasks in world model learning. Our experiments show that the base MBRL method equipped with HarmonyDream gains 10%-69% absolute performance boosts on visual robotic tasks and sets a new state-of-the-art result on the Atari 100K benchmark. Code is available at https://github.com/thuml/HarmonyDream.