← Back to Videos

RL Post-Training Generalizes Robot Policies Beyond Seen Tasks

Loading video

A new reinforcement-learning recipe improves generalist robot policies on trained tasks and transfers the gains to novel tasks not seen during training.

Robot Learning RL Post-Training Generalization VLA Reinforcement Learning

Category: research

Author: @MichalAKorniak

Date: 2026-11-05T00:00:00

Duration: 12.766s

Source: https://x.com/MichalAKorniak/status/2118387709349916895

VLA-Precision: one hour of real-robot online RL reaches 98.3% success

VLA-Precision: one hour of real-robot online RL reaches 98.3% success

@heetezition · 1.1k views · 2026-09-08

VLAReinforcement Learning

A-0 Trains Robot Policies on 16,000 Hours of Human Video

A-0 Trains Robot Policies on 16,000 Hours of Human Video

@yh_kris · 53 views · 2026-11-04

Human demonstrationsRobot Learning

Astra Edits Four Weights to Lift VLA Success

Astra Edits Four Weights to Lift VLA Success

@Lingxiao234 · 96 views · 2026-11-04

VLAWeight Editing

Same VLA, three data sources: Human Archive OTS wins on task success

Same VLA, three data sources: Human Archive OTS wins on task success

@babugi28 · 197 views · 2026-10-15

VLAPre-training Data

Long-WAM: Scaling World-Action Model Context to 19.2s in Real Time

Long-WAM: Scaling World-Action Model Context to 19.2s in Real Time

@AaronWeiHuang · 538 views · 2026-10-08

World-action modelsVLA

FLUX 3 Action: Open-Weights 7B World Action Model

FLUX 3 Action: Open-Weights 7B World Action Model

@HuggingPapers · 861 views · 2026-09-23

World-action modelsVLA

As an Amazon Associate, we earn from qualifying purchases.