PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space

Name: PearlVLA: Progressive Embodied Action-Plan Refinement in Latent Space
Uploaded: 2026-10-10T18:44:10.660274+00:00
Duration: 483 s
Description: PearlVLA proposes a framework that refines action plans inside a vision-language model's latent space for better action planning and low latency. It uses progressive, closed-loop refinement where a plan-conditioned query probes a latent world model to guide iterative improvements, showing state-of-the-art results on benchmarks.

PearlVLA proposes a framework that refines action plans inside a vision-language model's latent space for better action planning and low latency. It uses progressive, closed-loop refinement where a plan-conditioned query probes a latent world model to guide iterative improvements, showing state-of-the-art results on benchmarks.

Open the full episode