The Object Goal Navigation (ObjectNav) task challenges agents to locate a specified object in an unseen environment by imagining unobserved regions of the scene. Prior approaches rely on deterministic and discriminative models to complete semantic maps, overlooking the inherent uncertainty in indoor layouts and limiting their ability to generalize to unseen environments. In this work, we propose GOAL, a generative flow-based framework that models the semantic distribution of indoor environments by bridging observed regions with LLM-enriched full-scene semantic maps. During training, spatial priors inferred from large language models (LLMs) are encoded as two-dimensional Gaussian fields and injected into target maps, distilling rich contextual knowledge into the flow model and enabling more generalizable completions. Extensive experiments demonstrate that GOAL achieves state-of-the-art performance on MP3D and Gibson, and shows strong generalization in transfer settings to HM3D.

These skills are refined and updated through an iterative comparison strategy, enabling efficient adaptation to unseen environments.

We demonstrate the versatility of our approach for both fully data-driven and for physics-aware neural solvers.

Neural Information Processing Systems http://nips.cc/

Vision-and-Language Navigation requires the agent to follow language instructions to navigate through 3D environments.

Directional features As in previous work [3, 6, 10], we apply a 128-dimensional directional encoding byreplicating(cosθi,sinθi,cosφi,sinφi)by32times torepresent theorientation ofeach single-viewiwith respect to the agent's current orientation, whereθi andφi are the angles of the heading and elevation to that single-view. Replicating the encoding by 32 times does not enrich its information but makes its gradient 32 times larger during back-propagation. We suspect that this benefits the agent to learn about the action-related terms (e.g.

We incorporate an exogenous variable in the SCM that is learned and utilised for reasoning about interventionsintheobservation.

Aprominent challenge is to train an agent capable of generalising to new environments attest time, rather than one that simply memorises trajectories and visual details observed during training. We propose a new learning strategy that learns both from observations and generatedcounterfactual environments.