Solaris: Building a Multiplayer Video World Model in Minecraft

arXiv 26.2 2026 70.7 method, benchmark, system, application

TLDR

Solaris introduces a multiplayer video world model for Minecraft, enabling consistent multi-view simulation of multi-agent interactions.

Reasoning

The paper addresses a clear gap (single-agent video world models) with a novel multiplayer data collection system and staged training pipeline. Strengths include open-sourcing and a dedicated evaluation framework; weaknesses are domain specificity to Minecraft and potential scalability issues.

Read-first score

Read-first score 70.7, weighted from topical fit, citation, graph, method, reproducibility, and recency signals. Original total remains 56.

Recency 8%

100

Uses a gentle age decay so recent papers surface without erasing older foundations. 2026

Topical relevance 42%

Uses existing LLM keyword relevance scores normalized to 0-100. world model,world simulator,generative world model,interactive world model,video world model,world dynamics prediction,model-based reinforcement learning world model

Methodology quality 25%

Screens visible abstract and analysis fields for experiment, dataset, baseline, metric, and limitation evidence. markers=baseline,evaluation,result

Reproducibility 25%

Screens links and visible text for paper, code, dataset, artifact, and repository signals. pdf=True; code=False; dataset=False; markers=checkpoint,github

Field roles

FrontierMethodology anchor

Rank sensitivity

Stability: volatile; rank range: 104.

Keyword Scores

video world model

world model

generative world model

interactive world model

world dynamics prediction

world simulator

model-based reinforcement learning world model

Deep Analysis

Innovations

Multiplayer video world model that simulates consistent multi-view observations
Multiplayer data system for robust, continuous, and automated data collection in games like Minecraft
Staged training pipeline that progressively transitions from single-player to multiplayer modeling
Checkpointed Self Forcing, a memory-efficient Self Forcing variant for longer-horizon teacher

Methodology

Solaris uses a multiplayer data system to collect 12.64 million multiplayer frames from Minecraft. It is trained with a staged pipeline that progressively transitions from single-player to multiplayer modeling, combining bidirectional, causal, and Self Forcing training. In the final stage, Checkpointed Self Forcing is introduced as a memory-efficient variant.

Key Results

Solaris outperforms existing baselines on the proposed evaluation framework for multiplayer movement, memory, grounding, building, and view consistency.

Resources

Paper Project