paper/arxiv/2608.29828

Robotics Dataset

This research paper introduces SymVD, a distillation framework that transfers knowledge from a large Vision-Language-Action (VLA) teacher model to a compact student policy by exploiting geometric symmetries in robotic manipulation tasks, using an equivariant actor-critic architecture and adaptive weighting to improve sample efficiency and generalization.

Catalog metadata

Repository owner

Hyewon Choi, Donggyu Kim, Soojean Han

Source

Research paper

Task

SymVD: Symmetric Vision Language Action Distillation for Robot Manipulation

Cameras

0

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