paper/arxiv/2608.26673

Robotics Dataset

This research paper introduces PredVLA, a sub-million-parameter predictive-coding policy for language-conditioned robot manipulation, achieving strong performance on LIBERO benchmarks without robot-data pretraining.

Catalog metadata

Repository owner

Hiroki Sawada, Shunichi Kasahara

Source

Research paper

Task

PredVLA: Predictive Sensorimotor Modeling for Sub-Million-Parameter Robot Manipulation

Cameras

0

Browse the catalog to compare robots, tasks, sensors, and licenses. Sign in to open protected download and model links.