Skip to content
RobotWorld
← Back

RL2-VLA: adaptive RL compositional steering framework

Loading video
Loading video

RL2-VLA uncovers distinct test-time scaling laws for VLAs in OOD scenarios and proposes an adaptive RL compositional steering framework to steer pretrained VLAs beyond dominant failure modes.

VLARL²-VLA测试时缩放强化学习OOD引导框架具身AITwitter
Category: research
Author: @derektan1995
Date: 2026-08-05T00:00:00
Duration: 19.4s
Source: https://x.com/derektan1995/status/2085126005447090294
Reference: https://github.com/marmotlab/RL2-VLA

Related Videos

pi0: generalist vision-language-action model
0:22

pi0: generalist vision-language-action model

@Psalteric · 5 views · 2026-08-04
VLAπ0Physical Intelligence
Ego2Robot: egocentric videos as robot training data
2:47

Ego2Robot: egocentric videos as robot training data

@xiong_hui_chen · 3 views · 2026-08-06
VLAEgo2Robot数据管线
Zetta ζ: Closed-Loop Recovery Harness on Frozen VLA Policies
0:17

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

@techniahqrobot · 52 views · 2026-08-30
ZettaVLA失败恢复
Apple EgoDex: Wear a Vision Pro, Do Chores — Dexterous Robot Data Without Teleop
0:30

Apple EgoDex: Wear a Vision Pro, Do Chores — Dexterous Robot Data Without Teleop

@DataScienceHarp · 13 views · 2026-08-27
灵巧操作数据采集第一人称视频
EchoWM: A Drivable World Model, Not Just a Clip
1:00

EchoWM: A Drivable World Model, Not Just a Clip

@JinWeiyang18434 · 19 views · 2026-08-24
世界模型World ModelEchoWM
ADEPT: Sim Pre-training for Zero-Shot Visuo-Tactile Dexterity
3:08

ADEPT: Sim Pre-training for Zero-Shot Visuo-Tactile Dexterity

@jayjunleee · 7 views · 2026-08-20
ADEPT灵巧手灵巧操作