Skip to content
RobotWorld
← Back to Videos

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

Loading video
Loading video

AXIS lets volunteers teleoperate a simulated Franka in a browser across 207 manipulation tasks, collecting 50K+ imperfect crowd demos. Continual pretraining on them lifts π0.5 from 83.9% to 88.8% on LIBERO-Plus, still improving from 25% to 50% to 100% of the dataset: diverse human mistakes cancel out while useful task behavior survives.

Robot ManipulationImitation LearningData engineTeleoperationVLA
Category: arm
Author: @heetezition
Date: 2026-09-06T00:00:00
Duration: 135.517s
Source: https://x.com/heetezition/status/2096501152737120508
Reference: https://axisaiorg.github.io/AXIS-V1/

Related Videos

OneWM: World and Action Models That Strengthen Each Other
2:52

OneWM: World and Action Models That Strengthen Each Other

@Selen7005717917 · 73 views · 2026-10-24
World ModelsRobot Manipulation
ASPIRE: Model Frontier Meets Robot Cost
1:18

ASPIRE: Model Frontier Meets Robot Cost

@DrJimFan · 171 views · 2026-10-16
NVIDIAASPIRE
ASPIRE: Open-Source Robotic Skill Discovery
1:18

ASPIRE: Open-Source Robotic Skill Discovery

@guanzhi_wang · 153 views · 2026-10-16
NVIDIAASPIRE
MiniCPM-Robot: 1.5B model beats larger robot-arm models
0:27

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

@oliviscusAI · 1k views · 2026-09-04
Robot ManipulationVLA
3DPWM: Completed Point Clouds for Robot Planning
0:55

3DPWM: Completed Point Clouds for Robot Planning

@fuxinli2 · 120 views · 2026-10-20
3DPWMPoint Cloud Completion
Proxy Policy Steering: Test-Time Robot Policy Adaptation
0:43

Proxy Policy Steering: Test-Time Robot Policy Adaptation

@LeoKharon · 630 views · 2026-09-26
Proxy Policy SteeringRobot Policy Adaptation

As an Amazon Associate, we earn from qualifying purchases.