Skip to content
RobotWorld
← Back to Videos

SuperMimic Scales Humanoid Control with 1,000x Fewer GPU-Hours

Loading video
Loading video

SuperMimic trains humanoid control across 129,514 motions on one GPU in three hours, compared with 128 GPUs for seven days in the prior state of the art, cutting GPU-hours by more than 1,000x.

Humanoid ControlMotion ImitationSuperMimicMotion retargetingScalable Training
Category: humanoid
Author: @qiayuanliao
Date: 2026-10-21T00:00:00
Duration: 74.133s
Source: https://x.com/qiayuanliao/status/2112936035911135695
Reference: https://supermimic.github.io

Related Videos

HumanoidPose Estimates Full-Body Humanoid Pose from Video
1:17

HumanoidPose Estimates Full-Body Humanoid Pose from Video

@cs_jiaxi_jiang · 91 views · 2026-10-22
HumanoidPose estimation
TERRA: Reconstructing Terrains from Motion Data for Muscle-Actuated Control
3:00

TERRA: Reconstructing Terrains from Motion Data for Muscle-Actuated Control

@tokufxug · 128 views · 2026-10-18
Musculoskeletal ModelMotion retargeting
DexWeave: Learning Dexterous Humanoid Loco-Manipulation from Human Demonstrations
1:40

DexWeave: Learning Dexterous Humanoid Loco-Manipulation from Human Demonstrations

@ErenChenAI · 275 views · 2026-10-09
Mobile ManipulationHumanoid
DynaRetarget: Learning Humanoid Robot Motions from Human Demonstrations
0:25

DynaRetarget: Learning Humanoid Robot Motions from Human Demonstrations

@spaceandtech_ · 610 views · 2026-10-02
HumanoidMotion retargeting
ORCS Releases a Humanoid Whole-Body Controller Toolkit
0:14

ORCS Releases a Humanoid Whole-Body Controller Toolkit

@lok_s_h · 1.1k views · 2026-09-14
HumanoidWhole-Body Control
UMR: Unified Motion Retargeting via Learned Point Cloud Correspondence
2:03

UMR: Unified Motion Retargeting via Learned Point Cloud Correspondence

@Runyi_Ingrid_Yu · 720 views · 2026-09-11
HumanoidMotion retargeting

As an Amazon Associate, we earn from qualifying purchases.