DroneWAM: Efficient World Action Model for Drone Visual Navigation
DroneWAM predicts drone navigation futures in latent space, cutting rollout depth while improving accuracy.
DroneWAM is a JEPA-style world-action model for drone visual navigation that predicts future states in representation space rather than generating images. A pretrained Resampler compresses encoder tokens, and a preference-trained Gate allocates rollout depth by scene. On the new DroneNav-6D simulation, which includes RGB, 6-DoF trajectories, controls, and wind, adaptive rollout reduces average prediction depth from 8 to 4.58 while improving trajectory accuracy over compared methods.
- JEPA latent prediction avoids generating future images explicitly.
- A preference-trained gate adapts prediction depth to the scene.
- DroneNav-6D includes 6-DoF flight, controls, and randomized wind.
- Adaptive rollout cuts mean depth from 8 to 4.58 and improves accuracy.
From the original article
1 min read at huggingface.co
World-action models give visual navigation agents a way to anticipate how candidate actions will change future observations and to act from the predicted consequences. For drones, this capability must operate under tight accuracy and efficiency constraints. We present DroneWAM, an efficient world-action model for drone visual navigation.
By Liang Yao, Fan Liu, Hongbo Lu · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.