DashVMC enables real-time control for Geometry Dash gameplay
DashVMC: Real-Time Discrete World Model Control in Geometry Dash
Machine Learning
Summary
Playing fast video games like Geometry Dash requires quick decisions within tiny time windows. The authors developed DashVMC, a computer program that learns how the game world works by watching recorded gameplay for a few hours. This program can then predict what will happen next and decide the best actions to take, all in real time. They tested this by first copying human actions and then improving them without any new gameplay data, and it performed better on multiple levels. The system runs fast enough on a normal computer to control the game live.
What this means in practice
- •For video game developers: Integrate real-time learned world models for AI agents that control games with strict frame deadlines without live game interaction.
- •For robotics engineers: Use compact action-conditioned world models to control systems needing fast decision loops under strict timing constraints.
Authors
Florent Tariolle, Florian Yger
Abstract
World-model agents are usually evaluated in simulators that can wait for the policy; live games impose the opposite constraint, requiring capture, prediction, and action before the next frame. We present DashVMC, which learns a compact, action-conditioned world model from approximately two hours of recorded Geometry Dash gameplay. To test whether the learned dynamics are actionable, a controller is initialized by behavioural cloning (BC) and refined with Proximal Policy Optimization (PPO) entirely in frozen-model rollouts, without further interaction with the live game. Across three controller seeds, the refined policies survive longer than their BC initializations on all three official levels and a held-out community layout. At deployment, the baseline skips visual generation and sustains a 60-Hz capture-to-action loop on a consumer GPU. Action-conditioned continuations and rollout diagnostics show that the model remains useful for control despite imperfect long-horizon fidelity.