Skip to content
RobotWorld
← Back to Videos

Real-Time EXPO-FT: Reliable VLA Fine-Tuning

Loading video
Loading video

A lightweight edit policy corrects VLA action candidates using the latest observation, while value-based selection trains the policy; four dynamic real-robot tasks improve from 42% to 97% within 10 minutes of online data.

VLARobotic ManipulationReinforcement LearningReal-time controlStanford UniversityVision-language-action (VLA)Robot Manipulation
Category: arm
Author: @robotsdigest
Date: 2026-09-24T00:00:00
Duration: 59.875s
Source: https://x.com/robotsdigest/status/2103188670903923002
Reference: https://arxiv.org/abs/2609.18207

Related Videos

MiniCPM-Robot: 1.5B model beats larger robot-arm models
0:27

MiniCPM-Robot: 1.5B model beats larger robot-arm models

@oliviscusAI · 655 views · 2026-09-04
Robot ManipulationVLAOpenBMB
Splat-MOVER: A Robot That Updates Its 3D World After Every Move
3:31

Splat-MOVER: A Robot That Updates Its 3D World After Every Move

@Alacritic_Super · 229 views · 2026-09-20
3D Gaussian SplattingRobotic ManipulationGrasp Generation
TEMPO Gives a VLA a Short Memory for Dynamic Manipulation
1:13

TEMPO Gives a VLA a Short Memory for Dynamic Manipulation

@heetezition · 186 views · 2026-09-17
VLADynamic ManipulationTemporal Context
TEMPO: Adding Temporal Context to a Pretrained VLA for Dynamic Manipulation
0:24

TEMPO: Adding Temporal Context to a Pretrained VLA for Dynamic Manipulation

@techniahqrobot · 204 views · 2026-09-17
VLADynamic ManipulationTemporal Context
reBot Leader Arm Mirrors Teleoperation into Isaac Sim
0:59

reBot Leader Arm Mirrors Teleoperation into Isaac Sim

@kabilankb2003 · 352 views · 2026-09-14
Robot TeleoperationReal-to-SimIsaac Sim
AXIS: A Community-Driven Data Engine That Makes π0.5 Better
2:16

AXIS: A Community-Driven Data Engine That Makes π0.5 Better

@heetezition · 303 views · 2026-09-06
Robot ManipulationImitation LearningData engine