HTTP/2 301 server: GitHub.com content-type: text/html location: https://xiaoyilin-code.github.io/Dexflow_page/ x-github-request-id: D7AD:2F7ECD:A6C9E8:BB436B:695420CE accept-ranges: bytes age: 0 date: Tue, 30 Dec 2025 18:58:23 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210035-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767121103.868097,VS0,VE200 vary: Accept-Encoding x-fastly-request-id: bcbfd164e2e58ad8b3ec95d49ab05dccfe7e48ca content-length: 162 HTTP/2 200 server: GitHub.com content-type: text/html; charset=utf-8 x-origin-cache: HIT last-modified: Mon, 24 Mar 2025 13:03:13 GMT access-control-allow-origin: * strict-transport-security: max-age=31556952 etag: W/"67e15811-5957" expires: Tue, 30 Dec 2025 19:08:23 GMT cache-control: max-age=600 content-encoding: gzip x-proxy-cache: MISS x-github-request-id: 5ADA:2D64E0:A5B64A:BA2F29:695420CE accept-ranges: bytes age: 0 date: Tue, 30 Dec 2025 18:58:23 GMT via: 1.1 varnish x-served-by: cache-bom-vanm7210035-BOM x-cache: MISS x-cache-hits: 0 x-timer: S1767121103.081647,VS0,VE224 vary: Accept-Encoding x-fastly-request-id: b43bc741e5cf082acb5cc04385713ada4c1a9a61 content-length: 6293

DexFlow

Unified Framework for Dexterous Manipulation

Xiaoyi Lin¹, Kunpeng Yao², Lixin Xu³, Xueqiang Wang⁴, Xuetao Li¹, Yuchen Wang¹, Miao Li^4,†

IEEE/RSJ IROS 2025

Paper Code(coming soon) arXiv

Technical Overview

Abstract

We present DexFlow, a novel framework for dexterous hand pose retargeting and object interaction modeling. Our method combines multi-source human demonstration data with physics-based optimization to generate natural robotic manipulation sequences. The key innovation lies in our contact-aware refinement system that preserves interaction fidelity while ensuring temporal coherence across frames.

Pipeline Overview

Our proposed grasp retargeting framework comprises three main modules. First, the object segmented from the multi-frame MANO and object interaction sequence is scaled, and the human hand pose is retargeted to a robotic hand pose. Next, a double-threshold detection system extracts initial contact information between the retargeted hand and the object, which is then smoothed over adjacent frames and updated only if certain conditions are met. Finally, each finger is optimized in sequence, starting from the thumb and moving toward the pinky. At each stage of optimization, one finger is refined, and fingers without contact information, such as the index finger, are skipped, ensuring an efficient and accurate optimization procedure.

Interactive Demonstration

AR Mode

Human Demonstration

Retargeted from MoCap data

AR Mode

Robotic Execution

Optimized contact points

Result Demonstrations