Search papers, labs, and topics across Lattice.
LatentPilot introduces a novel VLN paradigm that leverages future visual observations during training to learn action-conditioned visual dynamics, without requiring future frames at inference. The method employs a flywheel-style training mechanism with on-policy trajectory collection and expert takeover to improve behavior distribution matching. By learning and propagating visual latent tokens across steps, LatentPilot enables the agent to "dream ahead" and reason about action consequences, achieving state-of-the-art results on R2R-CE, RxR-CE, and R2R-PE benchmarks, as well as demonstrating improved real-world performance.
VLN agents can now "dream ahead" by learning action-conditioned visual dynamics in a latent space, leading to SOTA results and improved real-world navigation.
Existing vision-and-language navigation (VLN) models primarily reason over past and current visual observations, while largely ignoring the future visual dynamics induced by actions. As a result, they often lack an effective understanding of the causal relationship between actions and how the visual world changes, limiting robust decision-making. Humans, in contrast, can imagine the near future by leveraging action-dynamics causality, which improves both environmental understanding and navigation choices. Inspired by this capability, we propose LatentPilot, a new paradigm that exploits future observations during training as a valuable data source to learn action-conditioned visual dynamics, while requiring no access to future frames at inference. Concretely, we propose a flywheel-style training mechanism that iteratively collects on-policy trajectories and retrains the model to better match the agent's behavior distribution, with an expert takeover triggered when the agent deviates excessively. LatentPilot further learns visual latent tokens without explicit supervision; these latent tokens attend globally in a continuous latent space and are carried across steps, serving as both the current output and the next input, thereby enabling the agent to dream ahead and reason about how actions will affect subsequent observations. Experiments on R2R-CE, RxR-CE, and R2R-PE benchmarks achieve new SOTA results, and real-robot tests across diverse environments demonstrate LatentPilot's superior understanding of environment-action dynamics in scene. Project page:https://abdd.top/latentpilot/
