The paper proposes using world models only during training to distill foresight into a fast navigation policy, rather than deploying them online. By combining a visual action prior with two task‑aligned world models—one for social dynamics and one for scene semantics—the authors score candidate actions based on predicted consequences and refine the policy through group‑relative preference distillation. The resulting policy, which runs in a single forward pass at deployment, outperforms reactive and learned baselines in safety, social coherence, and goal‑directedness across five held‑out scenarios, and observers rate its motion as more human‑like.
arXiv:2609.39235v1 Announce Type: cross Abstract: World models offer a promising way to help robots understand how the physical world evolves and plan complex behaviours through imagination. Yet exis...
arXiv:2606. 12550v1 Announce Type: cross Abstract: Open-world mapless navigation from sparse language instructions requires resolving underspecified goals and inferring which environmental cues are relevant for reaching the goal.
End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments. Its standard training recipe, however, is expensive across all existing paradigms: imi...
The paper introduces TEMPO, a temporal-distance planning objective that augments latent world models by learning a map of the frozen latent space where distances correspond to the number of environment steps between states. TEMPO blends this learned temporal distance into the planner’s cost, requiring no additional rewards or policies, and adds negligible computational overhead. Experiments on eleven simulated environments show that TEMPO consistently improves planner performance across all goal distances, notably raising LeWM’s success rate from 36% to 99% on the TwoRoom task.
TerraTransfer++ demonstrates a demonstration‑free, end‑to‑end autonomous driving framework that leverages self‑play in vectorized simulators to pretrain a policy. The method distills self‑play rollouts into a flow‑matching trajectory teacher and trains lightweight adapters on a frozen vision backbone, using action KL divergence, a batch‑relational low‑rank loss, and trajectory‑matching loss. The resulting closed‑loop policy surpasses all published imitation‑based methods on the HUGSim benchmark, while the planner achieves state‑of‑the‑art performance on NAVSIM navhard and navtest without any human demonstrations.
arXiv:2606.20627v2 Announce Type: replace Abstract: Joint-Embedding Predictive Architectures (JEPAs) enable agents to plan in latent space by imagining the outcomes of candidate actions, yet task spe...
FIRM-WM is a compact pixel world model that separates a goal‑comparable configuration from a 128‑dimensional dynamic fiber, enabling reward‑free visual planning from offline videos. It addresses two key mismatches: aligning planning states with goal images and reconciling factual trajectories with interventional sampling. In experiments, FIRM‑WM achieves high success rates on TwoRoom, Reacher, and OGBench‑Cube while using fewer parameters and faster planning times than prior models.
The paper introduces Dwell, a model‑based reinforcement learning framework that treats a pretrained vision‑language model (VLM) as a recurrent latent transition operator. Dwell enables the VLM to predict, adjust, and act within its own latent space, allowing the agent to learn from real interactions and latent imagination without reconstructing full futures. Experiments across five interactive visual environments show that Dwell outperforms other trainable methods while significantly reducing computational demands for imagination.
arXiv:2606. 17386v1 Announce Type: cross Abstract: End-to-end autonomous driving has achieved state-of-the-art performance on benchmarks and real-world deployments.
WALT introduces a method to align latent trajectories with pretrained driving world models, creating a compact generative trajectory space that preserves action-relevant semantics without altering the original model. The approach uses a dual-branch autoencoder to map raw waypoints into this latent space and transfers visual world knowledge into trajectory representations. Experiments on NAVSIM benchmarks show modest performance gains and a 30.5% reduction in planner FLOPs, indicating that maintaining world representations while extracting action-relevant information can improve trajectory planning efficiency.
arXiv:2609.38334v1 Announce Type: new Abstract: Large language models (LLMs) are increasingly deployed as agents for multi-step decision-making, yet transfer poorly to unseen environments. World-mode...
The paper introduces Imagine-then-Plan (ITP), a framework that lets agents learn by interacting with a learned world model to generate multi-step imagined trajectories. ITP features an adaptive lookahead mechanism that balances ultimate goals with task progress, producing richer signals about future outcomes. Experiments on various benchmarks show that ITP outperforms existing baselines, and analyses suggest the adaptive lookahead improves reasoning for complex tasks.