Odysseus: Scaling VLMs to 100+ Turn Decision-Making in Games via Reinforcement Learning

TL;DR

This paper introduces Odysseus, a framework that enhances vision-language models for long-horizon decision-making in video games using reinforcement learning, achieving significant progress and stability.

Contribution

It presents an adapted PPO algorithm with a turn-level critic and demonstrates the effectiveness of pretrained VLMs in long-term game environments, advancing RL training stability and sample efficiency.

Findings

Odysseus achieves at least 3 times more game progress than previous models.

Pretrained VLMs significantly improve sample efficiency and reduce manual action engineering.

The framework generalizes well across different game levels and settings.

Abstract

Given the rapidly growing capabilities of vision-language models (VLMs), extending them to interactive decision-making tasks such as video games has emerged as a promising frontier. However, existing approaches either rely on large-scale supervised fine-tuning (SFT) on human trajectories or apply reinforcement learning (RL) only in relatively short-horizon settings (typically around 20--30 turns). In this work, we study RL-based training of VLMs for long-horizon decision-making in Super Mario Land, a visually grounded environment requiring 100+ turns of interaction with coordinated perception, reasoning, and action. We begin with a systematic investigation of key algorithmic components and propose an adapted variant of PPO with a lightweight turn-level critic, which substantially improves training stability and sample efficiency over critic-free methods such as GRPO and Reinforce++. We further show that pretrained VLMs provide strong action priors, significantly improving sample efficiency during RL training and reducing the need for manual design choices such as action engineering, compared to classical deep RL trained from scratch. Building on these insights, we introduce Odysseus, an open training framework for VLM agents, achieving substantial gains across multiple levels of the game and at least 3 times average game progresses than frontier models. Moreover, the trained models exhibit consistent improvements under both in-game and cross-game generalization settings, while maintaining general-domain capabilities. Overall, our results identify key ingredients for making RL stable and effective in long-horizon, multi-modal settings, and provide practical guidance for developing VLMs as embodied agents.