Transformers have shown strong ability to model long-term dependencies and are increasingly adopted as world models in model-based reinforcement learning (RL) under partial observability. However, unlike natural language corpora, RL trajectories are sparse and reward-driven, making standard self-attention inefficient because it distributes weight uniformly across all past tokens rather than emphasizing the few transitions critical for control. To address this, we introduce structured inductive priors into the self-attention mechanism of the dynamics head: (i) per-head memory-length priors that constrain attention to task-specific windows, and (ii) distributional priors that learn smooth Gaussian weightings over past state-action pairs. We integrate these mechanisms into UniZero, a model-based RL agent with a Transformer-based world model that supports planning under partial observability. Experiments on the Atari 100k benchmark show that most efficiency gains arise from the Gaussian prior, which smoothly allocates attention to informative transitions, while memory-length priors often truncate useful signals with overly restrictive cut-offs. In particular, Gaussian Attention achieves a 77% relative improvement in mean human-normalized scores over UniZero. These findings suggest that in partially observable RL domains with non-stationary temporal dependencies, discrete memory windows are difficult to learn reliably, whereas smooth distributional priors flexibly adapt across horizons and yield more robust data efficiency. Overall, our results demonstrate that encoding structured temporal priors directly into self-attention improves the prioritization of informative histories for dynamics modeling under partial observability.

In heterogeneous multi-task decision-making, tasks not only exhibit diverse observation and action spaces but also vary substantially in their underlying complexities. While conventional multi-task world models like UniZero excel in single-task settings, we find that when handling a broad and diverse suite of tasks, gradient conflicts and the loss of model plasticity often constrain their sample efficiency. In this work, we address these challenges from two complementary perspectives: the single learning iteration and the overall learning process. First, to mitigate the gradient conflicts, we systematically investigate key architectural designs for extending UniZero. Our investigation identifies a Mixture-of-Experts (MoE) architecture as the most effective approach. We demonstrate, both theoretically and empirically, that this architecture alleviates gradient conflicts by routing task-specific representations to specialized sub-networks. This finding leads to our proposed model, \textit{ScaleZero}. Second, to dynamically allocate model capacity throughout the learning process, we introduce an online Dynamic Parameter Scaling (DPS) strategy. This strategy progressively integrates LoRA adapters in response to task-specific progress, enabling adaptive knowledge retention and parameter expansion. Evaluations on a diverse set of standard benchmarks (Atari, DMC, Jericho) demonstrate that ScaleZero, utilizing solely online reinforcement learning with one model, performs on par with specialized single-task agents. With the DPS strategy, it remains competitive while using just 71.5% of the environment interactions. These findings underscore the potential of ScaleZero for effective multi-task planning. Our code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.

Learning predictive world models is essential for enhancing the planning capabilities of reinforcement learning agents. Notably, the MuZero-style algorithms, based on the value equivalence principle and Monte Carlo Tree Search (MCTS), have achieved superhuman performance in various domains. However, in environments that require capturing long-term dependencies, MuZero's performance deteriorates rapidly. We identify that this is partially due to the \textit{entanglement} of latent representations with historical information, which results in incompatibility with the auxiliary self-supervised state regularization. To overcome this limitation, we present \textit{UniZero}, a novel approach that \textit{disentangles} latent states from implicit latent history using a transformer-based latent world model. By concurrently predicting latent dynamics and decision-oriented quantities conditioned on the learned latent history, UniZero enables joint optimization of the long-horizon world model and policy, facilitating broader and more efficient planning in latent space. We demonstrate that UniZero, even with single-frame inputs, matches or surpasses the performance of MuZero-style algorithms on the Atari 100k benchmark. Furthermore, it significantly outperforms prior baselines in benchmarks that require long-term memory. Lastly, we validate the effectiveness and scalability of our design choices through extensive ablation studies, visual analyses, and multi-task learning results. The code is available at \textcolor{magenta}{https://github.com/opendilab/LightZero}.