The Global Neural World Model: Spatially Grounded Discrete Topologies for Action-Conditioned Planning
This work addresses the problem of stable and generalizable world models for planning in reinforcement learning and robotics, offering a novel approach to avoid drift and learn causal structures without expert data.
The Global Neural World Model (GNWM) introduces a self-stabilizing framework that uses topological quantization on a discrete 2D grid to achieve action-conditioned planning without pixel-level reconstruction. It prevents manifold drift during autoregressive rollouts and learns generalized transition dynamics via maximum entropy exploration, validated across passive observation, active control, and abstract sequence tasks.
We present the Global Neural World Model (GNWM), a self-stabilizing framework that achieves topological quantization through balanced continuous entropy constraints. Operating as a continuous, action-conditioned Joint-Embedding Predictive Architecture (JEPA), the GNWM maps environments onto a discrete 2D grid, enforcing translational equivariance without pixel-level reconstruction. Our results show this architecture prevents manifold drift during autoregressive rollouts by using grid ``snapping'' as a native error-correction mechanism. Furthermore, by training via maximum entropy exploration (random walks), the model learns generalized transition dynamics rather than memorizing specific expert trajectories. We validate the GNWM across passive observation, active agent control, and abstract sequence regimes, demonstrating its capacity to act not just as a spatial physics simulator, but as a causal discovery model capable of organizing continuous, predictable concepts into structured topological maps.