Patch Policy: pretrained ViT beats OpenVLA-OFT with 0.7% params

Loading video

Patch Policy uses pretrained ViT plus a small transformer to beat OpenVLA-OFT with only 0.7% of its params, trainable on a single RTX 5090, demonstrating cable insertion at 2mm tolerance.

VLA Patch Policy ViT OpenVLA manipulation research Twitter

Category: research

Author: @jeffacce

Date: 2026-07-22T00:00:00

Duration: 21.883s

Source: https://x.com/jeffacce/status/2080017577749684718

Zetta ζ: Closed-Loop Recovery Harness on Frozen VLA Policies

Zetta ζ: Closed-Loop Recovery Harness on Frozen VLA Policies

@techniahqrobot · 52 views · 2026-08-30

ZettaVLA失败恢复

Patch Policy: small transformer outperforms large VLAs

Patch Policy: small transformer outperforms large VLAs

@LeoKharon · 4 views · 2026-08-12

Patch PolicyVLAtransformer

mimic-video Beats pi0 VLAs on LIBERO

mimic-video Beats pi0 VLAs on LIBERO

@cybernetic_lab · 3 views · 2026-07-29

VLAmimic-videoLIBERO

Imitation + reinforcement learning for robust towel folding

Imitation + reinforcement learning for robust towel folding

@SciRobotics · 4 views · 2026-07-22

模仿学习imitation learning强化学习

Agentic Real2Sim: reconstructing a physics simulator from a single video

Agentic Real2Sim: reconstructing a physics simulator from a single video

@robotsdigest · 3 views · 2026-07-22

world modelAgentic Real2Simphysics simulation

Flex-π: Multi-Stream World-Action Model

Flex-π: Multi-Stream World-Action Model

@GeYan_21 · 9 views · 2026-08-12

VLA世界模型WAM