Policy-Guided World Model Planning for Language-Conditioned Visual Navigation
Summary: arXiv:2603.25981v1 Announce Type: cross
Abstract: Navigating to a visually specified goal given natural language instructions remains a fundamental challenge in embodied AI. Existing approaches either rely on reactive policies that struggle with long-horizon planning, or employ world models that suffer from poor action initialization in high-dimensional spaces. We present PiJEPA, a two-stage framework that combines the strengths of learned navigation policies with latent world model planning for instruction-conditioned visual navigation.
In the first stage, we finetune an Octo-based generalist policy, augmented with a frozen pretrained vision encoder (DINOv2 or V-JEPA-2), on the CAST navigation dataset to produce an informed action distribution conditioned on the current observation and language instruction. In the second stage, we use this policy-derived distribution to warm-start Model Predictive Path Integral (MPPI) planning over a separately trained JEPA world model, which predicts future latent states in the embedding space of the same frozen encoder. By initializing the MPPI sampling distribution from the policy prior rather than from an uninformed Gaussian, our planner converges faster to high-quality action sequences that reach the goal.
We systematically study the effect of the vision encoder backbone, comparing DINOv2 and V-JEPA-2, across both the policy and world model components. Experiments on real-world navigation tasks demonstrate that PiJEPA significantly outperforms both standalone policy execution and uninformed world model planning, achieving improved goal-reaching accuracy and instruction-following fidelity.
Key Features of PiJEPA
- Two-Stage Framework: Combines learned navigation policies with latent world model planning.
- Fine-tuning of Generalist Policy: Utilizes an Octo-based generalist policy with a frozen pretrained vision encoder.
- Warm-Starting MPPI Planning: Enhances Model Predictive Path Integral planning with a policy-derived distribution.
- Encoder Backbone Comparison: Evaluates the performance of DINOv2 and V-JEPA-2 in both policy and world model components.
- Real-World Navigation Tasks: Demonstrates significant improvements in goal-reaching accuracy and instruction-following fidelity.
Conclusion
The development of PiJEPA represents a significant advancement in the field of embodied AI, particularly in the realm of visual navigation guided by natural language instructions. By effectively integrating learned policies and world models, PiJEPA not only addresses the limitations of existing approaches but also sets a new benchmark for future research in this area.