Meta tags:
description=MotionTrans: Human VR Data Enable Motion-Level Learning for Robotic Manipulation Policies;
keywords=Human data, Motion transfer, Cotraining, Policy learning;

Headings (most frequently used words):

motiontrans, human, robot, data, vr, motion, level, learning, demonstrations, shot, inference, results, on, real, tasks, enable, for, robotic, manipulation, policies, project, video, abstract, from, zero, few, collecting, and, acknowledgments, bibtex, hardware, system, based, capture, single, arm, platform, collection, demo, dataset, at, glance,

Text of the page (most frequently used words):
idx (56), pages (36), show (29), and (26), value (24), const (22), dot (22), classlist (22), addeventlistener (22), #motiontrans (22), human (21), foreach (20), dots (20), appendchild (18), active (18), data (18), robot (17), queryselectorall (16), function (16), for (14), video (14), setattribute (14), click (14), shot (13), pausehiddenpages (12), playvisiblepage (12), remove (12), aria (12), tasks (12), from (11), key (11), frag (10), pager (10), dotswrap (10), selected (10), add (10), motion (9), page (8), document (8), except (8), try (8), currenttime (8), catch (8), length (8), select (8), disabled (8), jump (7), sel (7), vla (7), task (7), fsselect (7), few (7), zsselect (7), zero (7), dotsfrag (6), let (6), array (6), false (6), true (6), slug (6), flash (6), learning (6), the (6), training (6), flashpage (6), vids (6), manipulation (5), policies (5), with (5), grid (4), createdocumentfragment (4), pause (4), play (4), prev (4), next (4), keydown (4), arrowleft (4), arrowright (4), gobtn (4), 1200 (4), change (4), level (4), robotic (4), pi0 (4), demonstrations (4), across (4), real (4), parseint (4), success (4), results (4), cotraining (4), university (4), card (3), jumpbyslug (3), index (3), zhou (3), wang (3), zhang (3), arxiv (3), our (3), their (3), are (3), support (3), box (3), collection (3), multi (3), state (3), fsgo (3), inference (3), rate (3), zsgo (3), its (3), this (3), code (3), push (2), getelementbyid (2), findindex (2), enable (2), author (2), yuan (2), chengbo (2), rui (2), liu (2), mengzhen (2), yingdong (2), shengjie (2), wen (2), chuan (2), shanghang (2), gao (2), yang (2), valuable (2), assistance (2), implementation (2), hardware (2), also (2), ruiqian (2), nai (2), huang (2), pour (2), bottle (2), towel (2), open (2), press (2), mango (2), bowl (2), bucket (2), transfer (2), hand (2), egocentric (2), unified (2), alignment (2), capture (2), system (2), based (2), innerhtml (2), createelement (2), span (2), classname (2), role (2), tab (2), controls (2), rvp (2), settimeout (2), scrollintoview (2), behavior (2), smooth (2), block (2), nearest (2), progress (2), score (2), quantitative (2), videos (2), left (2), right (2), comparison (2), between (2), deployable (2), transformation (2), weighted (2), use (2), while (2), such (2), datasets (2), remains (2), that (2), can (2), directly (2), motions (2), paper (2), potential (2), through (2), end (2), project (2), institute (2), information (2), shanghai (2), filter, indexcard, article, yuan2025motiontrans, title, journal, preprint, 2509, 17759, year, 2025, bibtex, would, like, express, sincere, gratitude, discussions, especially, grateful, his, thanks, extend, spiritai, inspirerobot, team, yankai, fanqi, lin, weirui, junmin, zhao, zhengrong, xue, jialei, haoxu, enshen, shuo, acknowledgments, milk, fold, panda, toy, bear, banana, plate, dice, lift, bypass, close, laptop, wipe, orange, stapler, bread, unplug, charger, collected, world, scenes, grouped, similar, skill, categories, cross, embodiment, 213, dataset, glance, during, log, keypoints, observations, textual, annotations, under, clock, making, downstream, straightforward, your, browser, does, not, tag, demo, synchronize, headset, controllers, view, cameras, trajectories, precise, time, single, arm, platform, collecting, demonstration, average, higher, better, red, yellow, blue, pink, middle, skills, via, action, space, scaling, bottleneck, imitation, leading, auxiliary, policy, other, aspects, image, language, understanding, may, learned, internet, acquiring, challenging, rich, diversity, behaviors, offers, resource, purpose, previous, works, using, bring, benefits, improving, robustness, efficiency, unclear, whether, realize, greatest, advantage, enabling, learn, new, completion, systematically, explore, introduce, framework, includes, simultaneously, notably, achieve, non, trivial, rates, manner, significantly, enhances, ablation, study, identify, factors, successful, these, findings, unlock, offering, insights, into, effective, all, model, weights, sourced, pretraining, finetuning, performance, more, than, strategy, pipeline, knowledge, abstract, teaser, pdf, interdisciplinary, sciences, tsinghua, zhi, laboratory, multimedia, processing, school, computer, science, peking, jiao, tong, wuhan, equal, contribution, corresponding, general, flow, related, research, skip, content,

Text of the page (random words):
motiontrans skip to content our related research general flow motiontrans motiontrans human vr data enable motion level learning for robotic manipulation policies chengbo yuan 1 2 rui zhou 5 mengzhen liu 3 yingdong hu 1 2 shengjie wang 1 2 li yi 1 2 chuan wen 4 shanghang zhang 3 yang gao 1 2 1 institute for interdisciplinary information sciences tsinghua university 2 shanghai qi zhi institute 3 state key laboratory of multimedia information processing school of computer science peking university 4 shanghai jiao tong university 5 wuhan university equal contribution corresponding author paper pdf arxiv code vla code datasets project video teaser project video abstract scaling real robot data is a key bottleneck in imitation learning leading to the use of auxiliary data for policy training while other aspects of robotic manipulation such as image or language understanding may be learned from internet based datasets acquiring motion knowledge remains challenging human data with its rich diversity of manipulation behaviors offers a valuable resource for this purpose while previous works show that using human data can bring benefits such as improving robustness and training efficiency it remains unclear whether it can realize its greatest advantage enabling robot policies to directly learn new motions for task completion in this paper we systematically explore this potential through multi task human robot cotraining we introduce motiontrans a framework that includes a data collection system a human data transformation pipeline and a weighted cotraining strategy by cotraining 30 human robot tasks simultaneously we directly transfer more than 10 motions from human data to deployable end to end robot policies notably 9 tasks achieve non trivial success rates in zero shot manner motiontrans also significantly enhances pretraining finetuning performance 40 success rate through ablation study we identify key factors for successful motion learning cotraining with robot data these findings unlock the potential of motion level learning from human data offering insights into its effective use for training robotic manipulation policies all data code and model weights are open sourced motion level learning from human demonstrations from vr human demonstrations to deployable zero few shot robot skills via a unified state action space human robot data transformation and weighted multi task co training motiontrans zero shot inference results on real robot left human data middle motiontrans dp right motiontrans pi0 vla jump to a task go videos quantitative results zero shot success rate sr across tasks comparison between motiontrans dp blue and motiontrans π0 vla pink zero shot motion progress score across tasks comparison between motiontrans dp red and motiontrans π0 vla yellow motiontrans dp motiontrans dp zero shot motiontrans pi0 vla motiontrans pi0 vla zero shot frag appendchild page pager appendchild frag dotswrap innerhtml const pages array from pager queryselectorall vp page pages foreach p k const dot document createelement span dot classname vp dot k 0 active dot setattribute role tab dot setattribute aria selected k 0 true false dot setattribute aria controls p id dot addeventlistener click show k dotswrap appendchild dot const dots array from dotswrap queryselectorall vp dot let idx 0 function pausehiddenpages except pages foreach pg k if k except pg queryselectorall video foreach v try v pause v currenttime 0 catch e function playvisiblepage i pages i queryselectorall video foreach v try v currenttime 0 v play catch e function show i pages idx classlist remove active dots idx classlist remove active dots idx setattribute aria selected false idx i pages length pages length pages idx classlist add active dots idx classlist add active dots idx setattribute aria selected true pausehiddenpages idx playvisiblepage idx prev addeventlistener click show idx 1 next addeventlistener click show idx 1 pager addeventlistener keydown e if e key arrowleft show idx 1 if e key arrowright show idx 1 pausehiddenpages 0 playvisiblepage 0 function flashpage i const vids pages i queryselectorall rvp video vids foreach v v classlist add flash settimeout vids foreach v v classlist remove flash 1200 pages i scrollintoview behavior smooth block nearest zsselect addeventlistener change zsgo disabled zsselect value if zsselect value const i parseint zsselect value 10 show i flashpage i zsgo addeventlistener click if zsselect value const i parseint zsselect value 10 show i flashpage i zsgo disabled zsselect value motiontrans few shot inference results on real robot left human data right motiontrans dp few shot jump to a task go videos quantitative results few shot motion progress score higher is better few shot success rate sr across tasks and average human data human demonstration robot inference few shot motiontrans dp motiontrans dp few shot frag appendchild page pager appendchild frag dotswrap innerhtml const pages array from pager queryselectorall vp page pages foreach p k const dot document createelement span dot classname vp dot k 0 active dot setattribute role tab dot setattribute aria selected k 0 true false dot setattribute aria controls p id dot addeventlistener click show k dotswrap appendchild dot const dots array from dotswrap queryselectorall vp dot let idx 0 function pausehiddenpages except pages foreach pg k if k except pg queryselectorall video foreach v try v pause v currenttime 0 catch e function playvisiblepage i pages i queryselectorall video foreach v try v currenttime 0 v play catch e function show i pages idx classlist remove active dots idx classlist remove active dots idx setattribute aria selected false idx i pages length pages length pages idx classlist add active dots idx classlist add active dots idx setattribute aria selected true pausehiddenpages idx playvisiblepage idx prev addeventlistener click show idx 1 next addeventlistener click show idx 1 pager addeventlistener keydown e if e key arrowleft show idx 1 if e key arrowright show idx 1 pausehiddenpages 0 playvisiblepage 0 function flashpage i const vids pages i queryselectorall rvp video vids foreach v v classlist add flash settimeout vids foreach v v classlist remove flash 1200 pages i scrollintoview behavior smooth block nearest fsselect addeventlistener change fsgo disabled fsselect value if fsselect value const i parseint fsselect value 10 show i flashpage i fsgo addeventlistener click if fsselect value const i parseint fsselect value 10 show i flashpage i fsgo disabled fsselect value collecting human and robot demonstrations hardware system vr based human data capture single arm robot platform we synchronize vr headset controllers and multi view cameras to capture 3d hand trajectories egocentric video and robot state with precise time alignment data collection demo your browser does not support the video tag during collection we log hand keypoints egocentric observations and textual annotations under a unified clock making downstream alignment and training straightforward motiontrans dataset at a glance we collected 3 213 demonstrations across 15 human tasks and 15 robot tasks in 10 real world scenes tasks are grouped by motion similar skill categories to support cross embodiment human robot co training and transfer human tasks jump to a task unplug charger bread bucket press stapler orange bucket wipe towel close laptop mango bowl bypass mango bowl lift press dice banana plate pour bottle toy bear box open box panda box fold towel pour milk bottle go grid appendchild card page appendchild grid pages push page const frag document createdocumentfragment pages foreach p frag appendchild p pager appendchild frag const dotsfrag document createdocumentfragment for let i 0 i show i dotsfrag appendchild dot dotswrap appendchild dotsfrag const dots array from dotswrap queryselectorall vp dot let idx 0 function pausehiddenpages except pages foreach pg k if k except pg queryselectorall video foreach v try v pause v currenttime 0 catch e function playvisiblepage i pages i queryselectorall video foreach v try v currenttime 0 v play catch e function show i pages idx classlist remove active dots idx classlist remove active dots idx setattribute aria selected false idx i pages length pages length pages idx classlist add active dots idx classlist add active dots idx setattribute aria selected true pausehiddenpages idx playvisiblepage idx prev addeventlistener click show idx 1 next addeventlistener click show idx 1 pager addeventlistener keydown e if e key arrowleft show idx 1 if e key arrowright show idx 1 pausehiddenpages 0 playvisiblepage 0 const select document getelementbyid ht filter const gobtn document getelementbyid ht go function jumpbyslug slug const index tasks findindex t t slug slug if index card classlist remove flash 1200 select addeventlistener change gobtn disabled select value if select value jumpbyslug select value gobtn addeventlistener click if select value jumpbyslug select value gobtn disabled select value robot tasks jump to a task go grid appendchild card page appendchild grid pages push page const frag document createdocumentfragment pages foreach p frag appendchild p pager appendchild frag const dotsfrag document createdocumentfragment for let i 0 i show i dotsfrag appendchild dot dotswrap appendchild dotsfrag const dots array from dotswrap queryselectorall vp dot let idx 0 function pausehiddenpages except pages foreach pg k if k except pg queryselectorall video foreach v try v pause v currenttime 0 catch e function playvisiblepage i pages i queryselectorall video foreach v try v currenttime 0 v play catch e function show i pages idx classlist remove active dots idx classlist remove active dots idx setattribute aria selected false idx i pages length pages length pages idx classlist add active dots idx classlist add active dots idx setattribute aria selected true pausehiddenpages idx playvisiblepage idx prev addeventlistener click show idx 1 next addeventlistener click show idx 1 pager addeventlistener keydown e if e key arrowleft show idx 1 if e key arrowright show idx 1 pausehiddenpages 0 playvisiblepage 0 function jump slug const index tasks findindex t t slug slug if indexcard classlist remove flash 1200 sel addeventlistener change go disabled sel value if sel value jump sel value go addeventlistener click if sel value jump sel value go disabled sel value acknowledgments we would like to express our sincere gratitude to shuo wang gu zhang enshen zhou haoxu huang jialei huang ruiqian nai zhengrong xue junmin zhao and weirui ye for their valuable discussions we are especially grateful to ruiqian nai and fanqi lin for their assistance with the implementation of pi0 vla and to yankai fu for his support with the hardware implementation our thanks also extend to the spiritai and inspirerobot team for their assistance bibtex article yuan2025motiontrans title motiontrans human vr data enable motion level learning for robotic manipulation policies author yuan chengbo and zhou rui and liu mengzhen and hu yingdong and wang shengjie and yi li and wen chuan and zhang shanghang and gao yang journal arxiv preprint arxiv 2509 17759 year 2025