Datasets:

velvet-pine-22
/

PULSE

volunteer string	scenario string	recording_id string	segment_idx int64	timestamp string	start_sec int64	end_sec int64	duration_sec int64	primitive string	hand string	is_left_hand int64	is_right_hand int64	object string	description string	description_variants string
v1	s1	v1s1	0	00:00:03-00:00:05	3	5	2	pull	right	0	1	laptop power adapter	Pull laptop power adapter with right hand.	["Pull laptop power adapter with right hand.", "Pulling laptop power adapter with right hand.", "Right hand pulls laptop power adapter.", "Laptop power adapter is pulled with right hand."]
v1	s1	v1s1	1	00:00:05-00:00:07	5	7	2	pull	right	0	1	laptop power adapter	Pull laptop power adapter with right hand.	["Pull laptop power adapter with right hand.", "Pulling laptop power adapter with right hand.", "Right hand pulls laptop power adapter.", "Laptop power adapter is pulled with right hand."]
v1	s1	v1s1	2	00:00:10-00:00:12	10	12	2	pick_up	both	1	1	paper	Pick up paper with both hands.	["Pick up paper with both hands.", "Picking up paper with both hands.", "Both hands pick up paper.", "Paper is picked up with both hands."]
v1	s1	v1s1	3	00:00:13-00:00:15	13	15	2	grasp	both	1	1	laptop	Grasp laptop with both hands.	["Grasp laptop with both hands.", "Grasping laptop with both hands.", "Both hands grasp laptop.", "Laptop is grasped with both hands."]
v1	s1	v1s1	4	00:00:15-00:00:17	15	17	2	move	both	1	1	laptop	Move laptop with both hands.	["Move laptop with both hands.", "Moving laptop with both hands.", "Both hands move laptop.", "Laptop is moved with both hands."]
v1	s1	v1s1	5	00:00:18-00:00:20	18	20	2	grasp	both	1	1	laptop	Grasp laptop with both hands.	["Grasp laptop with both hands.", "Grasping laptop with both hands.", "Both hands grasp laptop.", "Laptop is grasped with both hands."]
v1	s1	v1s1	6	00:00:24-00:00:26	24	26	2	grasp	both	1	1	tape	Grasp tape with both hands.	["Grasp tape with both hands.", "Grasping tape with both hands.", "Both hands grasp tape.", "Tape is grasped with both hands."]
v1	s1	v1s1	7	00:00:26-00:00:27	26	27	1	put_down	both	1	1	tape	Put down tape with both hands.	["Put down tape with both hands.", "Putting down tape with both hands.", "Both hands put down tape.", "Tape is put down with both hands."]
v1	s1	v1s1	8	00:00:27-00:00:29	27	29	2	pick_up	left	1	0	marker	Pick up marker with left hand.	["Pick up marker with left hand.", "Picking up marker with left hand.", "Left hand picks up marker.", "Marker is picked up with left hand."]
v1	s1	v1s1	9	00:00:30-00:00:32	30	32	2	place	right	0	1	marker	Place marker with right hand.	["Place marker with right hand.", "Placing marker with right hand.", "Right hand places marker.", "Marker is placed with right hand."]
v1	s1	v1s1	10	00:00:33-00:00:35	33	35	2	grasp	right	0	1	marker	Grasp marker with right hand.	["Grasp marker with right hand.", "Grasping marker with right hand.", "Right hand grasps marker.", "Marker is grasped with right hand."]
v1	s1	v1s1	11	00:00:36-00:00:38	36	38	2	place	left	1	0	marker	Place marker with left hand.	["Place marker with left hand.", "Placing marker with left hand.", "Left hand places marker.", "Marker is placed with left hand."]
v1	s1	v1s1	12	00:00:39-00:00:41	39	41	2	pick_up	both	1	1	wallet	Pick up wallet with both hands.	["Pick up wallet with both hands.", "Picking up wallet with both hands.", "Both hands pick up wallet.", "Wallet is picked up with both hands."]
v1	s1	v1s1	13	00:00:44-00:00:46	44	46	2	pick_up	right	0	1	laptop power adapter	Pick up laptop power adapter with right hand.	["Pick up laptop power adapter with right hand.", "Picking up laptop power adapter with right hand.", "Right hand picks up laptop power adapter.", "Laptop power adapter is picked up with right hand."]
v1	s1	v1s1	14	00:00:46-00:00:48	46	48	2	adjust	both	1	1	laptop power adapter	Adjust laptop power adapter with both hands.	["Adjust laptop power adapter with both hands.", "Adjusting laptop power adapter with both hands.", "Both hands adjust laptop power adapter.", "Laptop power adapter is adjusted with both hands."]
v1	s1	v1s1	15	00:00:49-00:00:51	49	51	2	hold	both	1	1	laptop power adapter	Hold laptop power adapter with both hands.	["Hold laptop power adapter with both hands.", "Holding laptop power adapter with both hands.", "Both hands hold laptop power adapter.", "Laptop power adapter is held with both hands."]
v1	s1	v1s1	16	00:00:52-00:00:54	52	54	2	put_down	both	1	1	laptop power adapter	Put down laptop power adapter with both hands.	["Put down laptop power adapter with both hands.", "Putting down laptop power adapter with both hands.", "Both hands put down laptop power adapter.", "Laptop power adapter is put down with both hands."]
v1	s1	v1s1	17	00:00:54-00:00:56	54	56	2	adjust	right	0	1	paper	Adjust paper with right hand.	["Adjust paper with right hand.", "Adjusting paper with right hand.", "Right hand adjusts paper.", "Paper is adjusted with right hand."]
v1	s1	v1s1	18	00:00:56-00:00:58	56	58	2	grasp	both	1	1	laptop power adapter	Grasp laptop power adapter with both hands.	["Grasp laptop power adapter with both hands.", "Grasping laptop power adapter with both hands.", "Both hands grasp laptop power adapter.", "Laptop power adapter is grasped with both hands."]
v1	s1	v1s1	19	00:00:59-00:01:01	59	61	2	adjust	both	1	1	laptop power adapter	Adjust laptop power adapter with both hands.	["Adjust laptop power adapter with both hands.", "Adjusting laptop power adapter with both hands.", "Both hands adjust laptop power adapter.", "Laptop power adapter is adjusted with both hands."]
v1	s1	v1s1	20	00:01:02-00:01:04	62	64	2	put_down	right	0	1	laptop power adapter	Put down laptop power adapter with right hand.	["Put down laptop power adapter with right hand.", "Putting down laptop power adapter with right hand.", "Right hand puts down laptop power adapter.", "Laptop power adapter is put down with right hand."]
v1	s1	v1s1	21	00:01:08-00:01:10	68	70	2	remove	both	1	1	laptop power adapter	Remove laptop power adapter with both hands.	["Remove laptop power adapter with both hands.", "Removing laptop power adapter with both hands.", "Both hands remove laptop power adapter.", "Laptop power adapter is removed with both hands."]
v1	s1	v1s1	22	00:01:18-00:01:20	78	80	2	grasp	both	1	1	laptop power adapter	Grasp laptop power adapter with both hands.	["Grasp laptop power adapter with both hands.", "Grasping laptop power adapter with both hands.", "Both hands grasp laptop power adapter.", "Laptop power adapter is grasped with both hands."]
v1	s1	v1s1	23	00:01:21-00:01:23	81	83	2	adjust	both	1	1	laptop power adapter	Adjust laptop power adapter with both hands.	["Adjust laptop power adapter with both hands.", "Adjusting laptop power adapter with both hands.", "Both hands adjust laptop power adapter.", "Laptop power adapter is adjusted with both hands."]
v1	s1	v1s1	24	00:01:24-00:01:26	84	86	2	put_down	right	0	1	laptop power adapter	Put down laptop power adapter with right hand.	["Put down laptop power adapter with right hand.", "Putting down laptop power adapter with right hand.", "Right hand puts down laptop power adapter.", "Laptop power adapter is put down with right hand."]
v1	s1	v1s1	25	00:01:27-00:01:29	87	89	2	grasp	left	1	0	laptop power adapter	Grasp laptop power adapter with left hand.	["Grasp laptop power adapter with left hand.", "Grasping laptop power adapter with left hand.", "Left hand grasps laptop power adapter.", "Laptop power adapter is grasped with left hand."]
v1	s1	v1s1	26	00:01:30-00:01:32	90	92	2	pick_up	both	1	1	paper	Pick up paper with both hands.	["Pick up paper with both hands.", "Picking up paper with both hands.", "Both hands pick up paper.", "Paper is picked up with both hands."]
v1	s1	v1s1	27	00:01:33-00:01:35	93	95	2	put_down	both	1	1	paper	Put down paper with both hands.	["Put down paper with both hands.", "Putting down paper with both hands.", "Both hands put down paper.", "Paper is put down with both hands."]
v1	s2	v1s2	0	00:00:00-00:00:05	0	5	5	open	both	1	1	box	Open box with both hands.	["Open box with both hands.", "Opening box with both hands.", "Both hands open box.", "Box is opened with both hands."]
v1	s2	v1s2	1	00:00:11-00:00:14	11	14	3	close	both	1	1	box	Close box with both hands.	["Close box with both hands.", "Closing box with both hands.", "Both hands close box.", "Box is closed with both hands."]
v1	s2	v1s2	2	00:00:14-00:00:16	14	16	2	open	both	1	1	box	Open box with both hands.	["Open box with both hands.", "Opening box with both hands.", "Both hands open box.", "Box is opened with both hands."]
v1	s2	v1s2	3	00:00:16-00:00:17	16	17	1	close	both	1	1	box	Close box with both hands.	["Close box with both hands.", "Closing box with both hands.", "Both hands close box.", "Box is closed with both hands."]
v1	s2	v1s2	4	00:00:17-00:00:20	17	20	3	push	both	1	1	box	Push box with both hands.	["Push box with both hands.", "Pushing box with both hands.", "Both hands push box.", "Box is pushed with both hands."]
v1	s2	v1s2	5	00:00:20-00:00:21	20	21	1	open	both	1	1	box	Open box with both hands.	["Open box with both hands.", "Opening box with both hands.", "Both hands open box.", "Box is opened with both hands."]
v1	s2	v1s2	6	00:00:21-00:00:24	21	24	3	close	both	1	1	box	Close box with both hands.	["Close box with both hands.", "Closing box with both hands.", "Both hands close box.", "Box is closed with both hands."]
v1	s2	v1s2	7	00:00:25-00:00:26	25	26	1	move	right	0	1	tape	Move tape with right hand.	["Move tape with right hand.", "Moving tape with right hand.", "Right hand moves tape.", "Tape is moved with right hand."]
v1	s2	v1s2	8	00:00:26-00:00:28	26	28	2	pull	both	1	1	tape	Pull tape with both hands.	["Pull tape with both hands.", "Pulling tape with both hands.", "Both hands pull tape.", "Tape is pulled with both hands."]
v1	s2	v1s2	9	00:00:28-00:00:29	28	29	1	place	both	1	1	tape	Place tape with both hands.	["Place tape with both hands.", "Placing tape with both hands.", "Both hands place tape.", "Tape is placed with both hands."]
v1	s2	v1s2	10	00:00:29-00:00:44	29	44	15	push	both	1	1	tape	Push tape with both hands.	["Push tape with both hands.", "Pushing tape with both hands.", "Both hands push tape.", "Tape is pushed with both hands."]
v1	s2	v1s2	11	00:00:45-00:00:48	45	48	3	push	both	1	1	box	Push box with both hands.	["Push box with both hands.", "Pushing box with both hands.", "Both hands push box.", "Box is pushed with both hands."]
v1	s2	v1s2	12	00:00:48-00:00:49	48	49	1	pick_up	both	1	1	box	Pick up box with both hands.	["Pick up box with both hands.", "Picking up box with both hands.", "Both hands pick up box.", "Box is picked up with both hands."]
v1	s2	v1s2	13	00:00:49-00:00:50	49	50	1	move	both	1	1	box	Move box with both hands.	["Move box with both hands.", "Moving box with both hands.", "Both hands move box.", "Box is moved with both hands."]
v1	s2	v1s2	14	00:00:50-00:00:52	50	52	2	place	both	1	1	box	Place box with both hands.	["Place box with both hands.", "Placing box with both hands.", "Both hands place box.", "Box is placed with both hands."]
v1	s3	v1s3	0	00:00:09-00:00:11	9	11	2	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	1	00:00:11-00:00:14	11	14	3	open	right	0	1	sealed jar	Open sealed jar with right hand.	["Open sealed jar with right hand.", "Opening sealed jar with right hand.", "Right hand opens sealed jar.", "Sealed jar is opened with right hand."]
v1	s3	v1s3	2	00:00:16-00:00:17	16	17	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	3	00:00:17-00:00:20	17	20	3	open	right	0	1	sealed jar	Open sealed jar with right hand.	["Open sealed jar with right hand.", "Opening sealed jar with right hand.", "Right hand opens sealed jar.", "Sealed jar is opened with right hand."]
v1	s3	v1s3	4	00:00:22-00:00:23	22	23	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	5	00:00:23-00:00:25	23	25	2	open	right	0	1	sealed jar	Open sealed jar with right hand.	["Open sealed jar with right hand.", "Opening sealed jar with right hand.", "Right hand opens sealed jar.", "Sealed jar is opened with right hand."]
v1	s3	v1s3	6	00:00:26-00:00:28	26	28	2	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	7	00:00:28-00:00:35	28	35	7	open	right	0	1	sealed jar	Open sealed jar with right hand.	["Open sealed jar with right hand.", "Opening sealed jar with right hand.", "Right hand opens sealed jar.", "Sealed jar is opened with right hand."]
v1	s3	v1s3	8	00:00:36-00:00:37	36	37	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	9	00:00:37-00:00:39	37	39	2	open	right	0	1	sealed jar	Open sealed jar with right hand.	["Open sealed jar with right hand.", "Opening sealed jar with right hand.", "Right hand opens sealed jar.", "Sealed jar is opened with right hand."]
v1	s3	v1s3	10	00:00:40-00:00:42	40	42	2	put_down	both	1	1	sealed jar	Put down sealed jar with both hands.	["Put down sealed jar with both hands.", "Putting down sealed jar with both hands.", "Both hands put down sealed jar.", "Sealed jar is put down with both hands."]
v1	s3	v1s3	11	00:00:42-00:00:43	42	43	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	12	00:00:43-00:00:47	43	47	4	remove	both	1	1	tea bag	Remove tea bag with both hands.	["Remove tea bag with both hands.", "Removing tea bag with both hands.", "Both hands remove tea bag.", "Tea bag is removed with both hands."]
v1	s3	v1s3	13	00:00:47-00:00:50	47	50	3	remove	both	1	1	tea bag	Remove tea bag with both hands.	["Remove tea bag with both hands.", "Removing tea bag with both hands.", "Both hands remove tea bag.", "Tea bag is removed with both hands."]
v1	s3	v1s3	14	00:00:50-00:00:52	50	52	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	15	00:00:53-00:00:54	53	54	1	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	16	00:00:54-00:00:55	54	55	1	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	17	00:00:56-00:00:57	56	57	1	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	18	00:00:57-00:00:58	57	58	1	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	19	00:00:58-00:01:00	58	60	2	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	20	00:01:00-00:01:02	60	62	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	21	00:01:02-00:01:04	62	64	2	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	22	00:01:04-00:01:06	64	66	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	23	00:01:06-00:01:08	66	68	2	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	24	00:01:08-00:01:10	68	70	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	25	00:01:10-00:01:12	70	72	2	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	26	00:01:12-00:01:14	72	74	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	27	00:01:14-00:01:16	74	76	2	grasp	right	0	1	tea bag	Grasp tea bag with right hand.	["Grasp tea bag with right hand.", "Grasping tea bag with right hand.", "Right hand grasps tea bag.", "Tea bag is grasped with right hand."]
v1	s3	v1s3	28	00:01:16-00:01:18	76	78	2	insert	right	0	1	tea bag	Insert tea bag with right hand.	["Insert tea bag with right hand.", "Inserting tea bag with right hand.", "Right hand inserts tea bag.", "Tea bag is inserted with right hand."]
v1	s3	v1s3	29	00:01:20-00:01:22	80	82	2	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	30	00:01:22-00:01:25	82	85	3	close	right	0	1	sealed jar	Close sealed jar with right hand.	["Close sealed jar with right hand.", "Closing sealed jar with right hand.", "Right hand closes sealed jar.", "Sealed jar is closed with right hand."]
v1	s3	v1s3	31	00:01:25-00:01:28	85	88	3	grasp	right	0	1	sealed jar	Grasp sealed jar with right hand.	["Grasp sealed jar with right hand.", "Grasping sealed jar with right hand.", "Right hand grasps sealed jar.", "Sealed jar is grasped with right hand."]
v1	s3	v1s3	32	00:01:28-00:01:30	88	90	2	grasp	left	1	0	sealed jar	Grasp sealed jar with left hand.	["Grasp sealed jar with left hand.", "Grasping sealed jar with left hand.", "Left hand grasps sealed jar.", "Sealed jar is grasped with left hand."]
v1	s3	v1s3	33	00:01:30-00:01:34	90	94	4	close	both	1	1	sealed jar	Close sealed jar with both hands.	["Close sealed jar with both hands.", "Closing sealed jar with both hands.", "Both hands close sealed jar.", "Sealed jar is closed with both hands."]
v1	s3	v1s3	34	00:01:35-00:01:36	95	96	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	35	00:01:36-00:01:42	96	102	6	close	right	0	1	sealed jar	Close sealed jar with right hand.	["Close sealed jar with right hand.", "Closing sealed jar with right hand.", "Right hand closes sealed jar.", "Sealed jar is closed with right hand."]
v1	s3	v1s3	36	00:01:43-00:01:44	103	104	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	37	00:01:44-00:01:49	104	109	5	close	right	0	1	sealed jar	Close sealed jar with right hand.	["Close sealed jar with right hand.", "Closing sealed jar with right hand.", "Right hand closes sealed jar.", "Sealed jar is closed with right hand."]
v1	s3	v1s3	38	00:01:50-00:01:51	110	111	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	39	00:01:51-00:01:56	111	116	5	close	right	0	1	sealed jar	Close sealed jar with right hand.	["Close sealed jar with right hand.", "Closing sealed jar with right hand.", "Right hand closes sealed jar.", "Sealed jar is closed with right hand."]
v1	s3	v1s3	40	00:01:56-00:01:57	116	117	1	grasp	both	1	1	sealed jar	Grasp sealed jar with both hands.	["Grasp sealed jar with both hands.", "Grasping sealed jar with both hands.", "Both hands grasp sealed jar.", "Sealed jar is grasped with both hands."]
v1	s3	v1s3	41	00:01:57-00:02:02	117	122	5	close	right	0	1	sealed jar	Close sealed jar with right hand.	["Close sealed jar with right hand.", "Closing sealed jar with right hand.", "Right hand closes sealed jar.", "Sealed jar is closed with right hand."]
v1	s3	v1s3	42	00:02:03-00:02:05	123	125	2	pick_up	left	1	0	cloth	Pick up cloth with left hand.	["Pick up cloth with left hand.", "Picking up cloth with left hand.", "Left hand picks up cloth.", "Cloth is picked up with left hand."]
v1	s3	v1s3	43	00:02:05-00:02:10	125	130	5	move	right	0	1	cloth	Move cloth with right hand.	["Move cloth with right hand.", "Moving cloth with right hand.", "Right hand moves cloth.", "Cloth is moved with right hand."]
v1	s3	v1s3	44	00:02:10-00:02:15	130	135	5	move	right	0	1	cloth	Move cloth with right hand.	["Move cloth with right hand.", "Moving cloth with right hand.", "Right hand moves cloth.", "Cloth is moved with right hand."]
v1	s3	v1s3	45	00:02:15-00:02:21	135	141	6	move	right	0	1	cloth	Move cloth with right hand.	["Move cloth with right hand.", "Moving cloth with right hand.", "Right hand moves cloth.", "Cloth is moved with right hand."]
v1	s3	v1s3	46	00:02:21-00:02:26	141	146	5	move	right	0	1	cloth	Move cloth with right hand.	["Move cloth with right hand.", "Moving cloth with right hand.", "Right hand moves cloth.", "Cloth is moved with right hand."]
v1	s3	v1s3	47	00:02:26-00:02:32	146	152	6	move	right	0	1	cloth	Move cloth with right hand.	["Move cloth with right hand.", "Moving cloth with right hand.", "Right hand moves cloth.", "Cloth is moved with right hand."]
v1	s3	v1s3	48	00:02:32-00:02:33	152	153	1	put_down	right	0	1	cloth	Put down cloth with right hand.	["Put down cloth with right hand.", "Putting down cloth with right hand.", "Right hand puts down cloth.", "Cloth is put down with right hand."]
v1	s3	v1s3	49	00:02:33-00:02:35	153	155	2	pick_up	both	1	1	sealed jar	Pick up sealed jar with both hands.	["Pick up sealed jar with both hands.", "Picking up sealed jar with both hands.", "Both hands pick up sealed jar.", "Sealed jar is picked up with both hands."]
v1	s3	v1s3	50	00:02:35-00:02:36	155	156	1	move	both	1	1	sealed jar	Move sealed jar with both hands.	["Move sealed jar with both hands.", "Moving sealed jar with both hands.", "Both hands move sealed jar.", "Sealed jar is moved with both hands."]
v1	s3	v1s3	51	00:02:36-00:02:37	156	157	1	place	both	1	1	sealed jar	Place sealed jar with both hands.	["Place sealed jar with both hands.", "Placing sealed jar with both hands.", "Both hands place sealed jar.", "Sealed jar is placed with both hands."]
v1	s3	v1s3	52	00:02:38-00:02:40	158	160	2	pick_up	both	1	1	sealed jar	Pick up sealed jar with both hands.	["Pick up sealed jar with both hands.", "Picking up sealed jar with both hands.", "Both hands pick up sealed jar.", "Sealed jar is picked up with both hands."]
v1	s3	v1s3	53	00:02:40-00:02:42	160	162	2	place	both	1	1	sealed jar	Place sealed jar with both hands.	["Place sealed jar with both hands.", "Placing sealed jar with both hands.", "Both hands place sealed jar.", "Sealed jar is placed with both hands."]
v1	s3	v1s3	54	00:02:46-00:02:48	166	168	2	place	left	1	0	sealed jar	Place sealed jar with left hand.	["Place sealed jar with left hand.", "Placing sealed jar with left hand.", "Left hand places sealed jar.", "Sealed jar is placed with left hand."]
v1	s4	v1s4	0	00:00:02-00:00:03	2	3	1	grasp	right	0	1	rice bowl	Grasp rice bowl with right hand.	["Grasp rice bowl with right hand.", "Grasping rice bowl with right hand.", "Right hand grasps rice bowl.", "Rice bowl is grasped with right hand."]
v1	s4	v1s4	1	00:00:03-00:00:04	3	4	1	grasp	left	1	0	rice bowl	Grasp rice bowl with left hand.	["Grasp rice bowl with left hand.", "Grasping rice bowl with left hand.", "Left hand grasps rice bowl.", "Rice bowl is grasped with left hand."]

End of preview. Expand in Data Studio

PULSE: A Synchronized Five-Modality Dataset for Sensorimotor Coordination in Long-Horizon Daily Activities

MoCap · EMG · Eye Tracking · IMU · Finger-level Pressure — time-aligned on a common 100 Hz timebase

NeurIPS 2026 Evaluations and Datasets Track (Spotlight) · CC BY-NC 4.0 · DOI 10.57967/hf/11314 · Paper (OpenReview)

At a glance

40	9	5	7,356
volunteers	scenarios (S1–S8 + S9 motion primitives)	modalities @ 100 Hz	dense action segments
337 total recordings (304 task + 33 S9)	~9.6 h total (S1–S8: ~6.8 h, S9: ~2.8 h)	common 100 Hz timebase	7 benchmark tasks (T1–T7)
282 annotated	36 annotated volunteers	18 motor primitives observed	54 unique objects

Release. Companion code (benchmark pipeline, 5-fold split files, VLM annotation prompt) is at velvet-pine-22/PULSE-code, a single-recording sample at velvet-pine-22/PULSE-sample, and Croissant metadata with RAI fields in croissant.json. The dataset is archived under the permanent DOI 10.57967/hf/11314. Contact and maintenance: see Contact and maintenance.

Tip. The Dataset Viewer at the top of this page offers the emg / mocap / pressure / action_segments / s9_primitive_segments subsets via the dropdown; hover any column to see its per-value distribution. The action_segments subset is the flat per-segment table (7,356 rows): hover the primitive, hand, or object columns to browse the label distributions. EyeTrack and IMU are not offered as viewer subsets because their column sets differ between recordings (see Column notes); load them per recording with pandas as shown below. The static figures further down summarise cross-tab statistics (per-modality coverage, per-volunteer matrix) the column-wise viewer cannot compute.

Quick start

Load EMG, MoCap, pressure or the label tables through datasets:

from datasets import load_dataset

emg = load_dataset("velvet-pine-22/PULSE", "emg", split="train")
# Available configs: emg, mocap, pressure, action_segments, s9_primitive_segments.
# (volunteer, scenario) is recovered from the source file path; the
# per-modality CSVs themselves do not carry id columns.

# Flat action-segments table (also browsable in the Dataset Viewer above):
seg = load_dataset("velvet-pine-22/PULSE", "action_segments", split="train")
# 7,356 rows × 15 columns: volunteer, scenario, primitive, hand, object, ...

Load any modality of one recording with pandas (recommended for EyeTrack and IMU, whose column sets vary between recordings, and for MoCap when you need the exact marker set of a recording):

import pandas as pd
from huggingface_hub import hf_hub_download

def load(volunteer, scenario, modality):
    path = hf_hub_download(
        "velvet-pine-22/PULSE",
        f"data/{volunteer}/{scenario}/aligned_{modality}_100hz.csv",
        repo_type="dataset",
    )
    return pd.read_csv(path)

imu = load("v30", "s1", "imu")        # 181 columns: time + 10 units × 18
eye = load("v30", "s1", "eyetrack")   # 26 columns: time, UTC + 24 features
# metadata/recordings.csv lists which modalities exist for each recording.

Inspect one recording on disk:

data/v1/s1/
├── aligned_emg_100hz.csv
├── aligned_eyetrack_100hz.csv
├── aligned_imu_100hz.csv
├── aligned_mocap_100hz.csv
├── aligned_pressure_100hz.csv
├── aligned_myo_pose_100hz.csv     # auxiliary (Myo pose class)
├── aligned_myo_quat_100hz.csv     # auxiliary (Myo orientation quaternion)
├── alignment_metadata.json
├── raw/                           # opt-in: Qualisys skeleton TSV (~30–40 MB)
│   └── aligned_v1s1_s_Q.tsv
└── videos/                        # opt-in: first-person video pair (~0.2 GB per recording;
                                   # no audio; incidentally captured faces pixelated)
    ├── scene_cam.mp4
    └── eye_tracking_overlay.mp4

Want only the benchmark streams? Skip the heavy bits at download:

hf download velvet-pine-22/PULSE --repo-type dataset \
    --exclude 'data/v*/s*/raw/**' \
    --exclude 'data/v*/s*/*_s_Q.tsv' \
    --exclude 'data/v*/s*/videos/**'
# ~14 GB instead of ~85 GB

(The second pattern catches the 10 Qualisys TSVs that sit next to the CSVs instead of under raw/; see Known issues.)

Dataset statistics

Modality availability

Realistic missing-modality pattern across the 304 S1–S8 task recordings: EMG (99.3%), IMU (98.4%), EyeTrack (92.8%), and Scene Cam (92.1%, 280/304) are present on >90% of task recordings; MoCap (80.9%) and Pressure (69.7%) are the two limiting modalities. All five sensors are simultaneously available on 198 / 304 task recordings (65.1%) — drives the T4 missing-modality benchmark. (The gaze-overlay video exists for 281 task recordings: v19/s5 has the overlay but no scene-camera file.)

Recordings and durations

Scenarios are well-balanced (37–40 recordings each for S1–S8, plus 33 S9 recordings). Recording length is right-skewed with a long tail of sessions up to ~9 minutes; median 77 s, mean 104 s.

Recordings per scenario and recording-duration histogram

Per-volunteer modality coverage and label distributions

The matrix below shows the missing-modality structure is systematic, not random: early volunteers (v1–v22) have intermittent MoCap and Pressure coverage; from v23 onwards almost all five modalities are present on every recording. Bimanual segments dominate (50%), and the manipulated-object distribution has a heavy head (top 8 of 54 objects account for ~51% of all segments).

Per-volunteer modality matrix; segments by hand; top manipulated objects

Motor-primitive distribution

7,356 dense segments covering all 18 motor primitives of the taxonomy (stabilize at 0.1% and support, with a single segment, are the rarest). The top three primitives — grasp / move / place — together account for 59% of all segments, reflecting the manipulation-heavy nature of daily activities.

Distribution of motor primitives across 7,356 dense action segments

The five modalities

Modality	Sensor	Released columns	Notes
MoCap	Qualisys optical	`aligned_mocap_100hz.csv`: 52 optical markers, each with X/Y/Z position (mm, laboratory frame) and the Qualisys marker `Type` (e.g. `Measured`) → 211 columns (`time`, `Frame`, `Time` + 52 × 4). `raw/aligned_vs_s_Q.tsv`: Qualisys skeleton with 56 segments (position + quaternion, 7 values each) and 10 fingertip points (3 values each) → 422 values plus `Frame`, `Time`	Positions are absolute. Hip-relative coordinates, velocities and the 21-joint hand used by T7 are computed by PULSE-code, not stored in the release
EMG	Myo Armband	8 channels (`emg_1`–`emg_8`): raw signed Myo signal (~200 Hz native) linearly interpolated to 100 Hz; no filtering or rectification	Right forearm for every participant (including the 14 left-handed participants)
EyeTrack	Dikablis Glasses 3	24 numeric features plus `time` and `UTC`: combined pupil X/Y; per eye pupil X/Y/area/height/width, fixation flag and duration, saccade flag, duration and angle; gaze X/Y in scene-camera coordinates	`v1/s1` (right eye only) and `v14/s8` (left eye only) lack one eye's 10 channels and are excluded from EyeTrack benchmarks; some recordings carry extra columns (Column notes)
IMU	10 wearable units (`WT0`–`WT9`)	180 values = 18 per unit: acceleration X/Y/Z (g), angular velocity X/Y/Z (°/s), angle X/Y/Z (°), magnetic field X/Y/Z (µT), quaternion (4), temperature (°C), battery level; plus `time`	Body-distributed. `v5` has only 4 units (`WT0`–`WT3`, 73 columns). Column names are partly Chinese, see Column notes
Pressure	Finger-level pressure array	50 channels (`R1(g)`–`R25(g)`, `L1(g)`–`L25(g)`; 25 per hand), in grams	Quantitative grip force

Auxiliary videos — each recording with video has up to two files: videos/scene_cam.mp4, the first-person scene-camera video of the Dikablis Glasses 3 (1920×1080, 30 fps), and videos/eye_tracking_overlay.mp4, the same view with the recorded gaze cursor drawn on it (1920×1080, kept at its recorded, variable frame rate: median ≈ 88 fps, range 30–90 fps). 311 scene-camera and 311 overlay files (63.1 GB) cover 312 recordings; v19/s5 has only the overlay and v15/s9 only the scene-camera file. Released videos contain no audio track, and the face of any incidentally captured laboratory member is pixelated. No eye-camera or third-person video is released. Videos of v5/s8 and v18/s2 are not released (see Known issues). The videos are privacy-sensitive auxiliary material and are not part of the T1–T7 benchmark protocol.

Synchronization — All streams are resampled onto a common 100 Hz timebase and placed on it from each device's recorded start timestamp; no hardware trigger is used. The aligned window is the overlap of all available streams (per-stream start and end times, and per-recording exceptions, are recorded in each recording's alignment_metadata.json).

Column notes

IMU column names follow the sensor software and are partly Chinese: WTk_加速度{X,Y,Z}_g = acceleration (g), WTk_角速度{X,Y,Z}_s = angular velocity (°/s), WTk_角度{X,Y,Z} = angle (°), WTk_磁场{X,Y,Z}_uT = magnetic field (µT), WTk_四元数{0..3} = quaternion, WTk_温度_C = temperature (°C), WTk_电量 = battery level.
metadata/modality_coverage.xlsx headers: 第一人称视频 = first-person scene-camera video (scene_cam.mp4), 注视点 = eye tracking, 压力信息 = pressure, 动捕 = MoCap, 时长(秒) = duration (s).
EyeTrack column sets: 291 of the 313 files have the 26 columns described above (column names are prefixed Dikablis Glasses 3_). v1/s1 and v14/s8 have 16 columns (one eye missing); v13/s2 and v13/s3 have 4 additional Index of Cognitive Activity columns (30); 18 recordings have 10 additional marker-coordinate columns (36) from the Dikablis marker-detection export.
MoCap column sets: 229 of the 246 files have the 211 columns described above. v2/s2, v2/s6, v3/s1, v3/s2, v3/s5, v3/s6, v13/s2 and v13/s3 have only 38 markers (155 columns); v11/s6 lacks the Type column of 13 markers (198 columns); a few further recordings list the same columns in a different order. Select columns by name.

Annotations

Two-level scheme. The dense segments were annotated by vision-language models (VLMs), and every segment was then confirmed or corrected by a human, so all released segments are human-reviewed; see Annotation provenance below.

L1 — scenario label (one of 8 + S9 motion-primitive protocol) per recording.
L2 — dense action segments (mean 2.2 s, median 2 s; boundaries at 1 s resolution). Each segment carries:
- motor primitive (taxonomy of 18; all 18 observed in this release)
- hand (left / right / both)
- manipulated object (scene-specific whitelist; 54 distinct objects observed)
- natural-language description with 4 paraphrased variants for language-grounded use

Annotations are released in two complementary forms:

annotations/v*/s*.json — one nested JSON per recording (canonical structure, referenced by croissant.json).
annotations_flat/segments.csv — flat table with one row per segment (used by the Dataset Viewer action_segments subset; 7,356 rows × 15 columns).

Per-volunteer × scenario segment counts are in metadata/annotations/segment_counts.{csv,xlsx}.

Annotation provenance

Action segments were produced in two steps. (1) VLM annotation: vision-language models segmented each egocentric scene-camera video under a scene-constrained prompt with the fixed 18-primitive vocabulary and per-scenario object lists (the prompt template is published in the prompts/ folder of PULSE-code). (2) Human review: each annotated recording was then checked segment by segment against its egocentric video with gaze overlay, and every segment was confirmed or corrected (editing, adding or deleting segments where needed). During review, nine objects missing from the per-scenario lists were added to the object vocabulary; the primitive vocabulary was not changed.

All released action segments have therefore been confirmed or corrected by a human: every segment of every annotated recording went through this review, except v40/s3, whose egocentric view mostly misses the hands and which the authors re-annotated from third-person video and motion-capture hand trajectories. v18/s2 carries no action labels because its egocentric video froze (see Known issues); it has no file under annotations/. The release contains 7,356 segments in 282 recordings.

S9 motion-primitive labels

30 of the 33 S9 recordings carry primitive-level labels (454 segments; v8/s9 and v10/s9 have no egocentric video and v14/s9 less than one second, so they are not labelled). They use the same file format (annotations/v*/s9.json, flat table annotations_flat/segments_s9.csv, Dataset Viewer subset s9_primitive_segments) but a different, closed vocabulary of 16 composite primitives that the volunteers performed on cue: pick_up_object, put_down_object, move_object, transfer_between_hands, open_lid, close_lid, pour_liquid, fold_clothing, stack_objects, wipe_surface, dump_trash, tear_tape, wrap_object, plug_in_power, stir, cut. These labels are kept apart from the S1–S8 segments: they are not counted in the 7,356 dense segments above, and has_dense_annotations in metadata/recordings.csv refers to the S1–S8 labels only. S9 labels went through the same two steps (VLM annotation, then human review of every segment). For v31/s9, v34/s9 and v38/s9 the human review also used a third-person reference video that is not released. v35/s9 is short (its egocentric video, and hence the aligned recording, ends after 130 s) and has only four segments.

Seven benchmark tasks

	Task	Headline modality use
T1	Scene recognition (8-way)	All five
T2	Fine-grained action recognition (motor primitive × object × hand)	All five
T3	Grasp onset anticipation	EMG + IMU lead
T4	Missing-modality robustness	Trained on all, evaluated under modality dropout
T5	Tactile-driven grasp state recognition	Pressure-centered sensorimotor windows
T6	Cross-modal pressure prediction	EMG / hand MoCap → pressure
T7	EMG-to-hand-pose reconstruction (21-joint right hand, MPJPE)	EMG → hand MoCap

DailyActFormer is provided as a strong T2 baseline: it has the highest mean headline score, but margins over the strongest published baselines are small and confidence intervals overlap. T3 and T6 are intended as open challenge tasks rather than solved benchmarks. Full results, ablations, and the grasp-phase timing analysis are in the accompanying paper (see Citation).

Evaluation protocol (volunteer-stratified 5-fold cross-validation)

All benchmark results in the paper use volunteer-stratified 5-fold cross-validation: every volunteer of a pool is in exactly one test fold, model selection uses an inner-validation group of training volunteers, and each fold is run with 3 seeds. The fold files are released in the code repository velvet-pine-22/PULSE-code:

Fold file	Volunteers	Used by
`splits/pool33.json`	33	T1, T4, T7
`splits/pool27.json`	27	T2, T3, T5, T6

Each fold file lists the pool, and for every fold the test, inner-validation and training volunteers.

Legacy split column. metadata/recordings.csv keeps a split column (train 242 recordings, test 40 recordings from v14, v30, v34, v38, v41, unannotated 55). It is a fixed subject-independent split from an earlier release and is not the protocol behind the paper's results; use the fold files above to reproduce them.

40 volunteers, IDs v1–v41 with v7 not assigned. S9 motion-primitive recordings are present for 33 volunteers and have no MoCap; 31 of the 33 have the scene-camera video (none for v8/s9, v10/s9) and 30 the gaze overlay (v15/s9 has the scene-camera video only).

Known issues

v18/s2 – frozen egocentric video, videos not released. The scene camera dropped out ~15.5 s into the raw recording; the trimmed egocentric and gaze-overlay videos are frozen after 6.5 s, so neither video is part of the release (only the sensor streams are). No action segments are released for this recording (no annotations/v18/s2.json; has_dense_annotations = 0 in metadata/recordings.csv). This does not mean that no actions occurred. All sensor streams are unaffected.
v5/s8 – videos not released. The scene camera ran ~21 min out of sync with the sensor streams in this recording, so neither the ego-centric nor the gaze-overlay video is part of the release; all sensor streams and annotations are unaffected.
v19/s5, v15/s9 – only one of the two videos. v19/s5 has only the gaze-overlay video (no scene_cam.mp4); v15/s9 has only scene_cam.mp4 (no gaze-overlay video). In metadata/recordings.csv, scene_video refers to scene_cam.mp4.
v36/s9, v11/s9 – scene video ends early. The scene-camera stream is ~152 s (v36/s9) and ~15 s (v11/s9) shorter than the recording, so the end of the gaze-overlay video is a frozen frame; S9 labels are given only for the part with a moving image.
v20/s2 – corrupted IMU units. In aligned_imu_100hz.csv, units WT4, WT5 and WT7 contain values that grow linearly over the recording to physically implausible magnitudes (up to ~7×10^5), and WT6 is frozen (zero angular velocity, constant orientation and magnetic field). Exclude these units from IMU analyses of this recording.
v5 – 4 IMU units. All IMU files of v5 contain only units WT0–WT3 (73 columns).
v17/s6 – skeleton only. raw/aligned_v17s6_s_Q.tsv is released, but there is no aligned_mocap_100hz.csv; metadata/recordings.csv lists MoCap as missing for this recording.
Qualisys TSVs outside raw/. For v1/s5, v2/s2, v2/s6, v3/s1, v3/s2, v3/s5, v3/s6, v11/s8, v13/s2 and v13/s3 the skeleton TSV sits next to the CSVs (data/vX/sY/aligned_vXsY_s_Q.tsv) instead of under raw/.
Varying column sets. EyeTrack, IMU and MoCap files do not share one schema across all recordings; see Column notes. Select columns by name.
Released MP4s use edit lists. Every released video keeps the MP4 edit list from trimming; decoders that ignore edit lists may return up to several seconds of extra footage at the start.
Manually trimmed recordings. v32/s1 (first 113 s) and v30/s6 (first 137 s) were trimmed to remove an aborted first attempt with the experimenter in view; see manual_trim in their alignment_metadata.json.
Placeholder rows. metadata/recordings.csv lists v4/s1 and v4/s4 with zero duration and no data; the 304 task recordings counted above include them (302 contain data).
Partial label coverage. Segments cover ~73% of the annotated recording time; intervals in which the hands or the manipulated object are out of view are left unlabelled. Boundaries have 1 s resolution.
Missing channels. 36 of the 282 annotated recordings have no MoCap stream; v14 lacks MoCap and pressure; v1/s1 and v14/s8 lack one eye's pupil-size and event channels.
metadata/batch_alignment_summary.json is a legacy log. It records one alignment batch (27 recordings of v3, v5, v9) and is partly outdated (e.g. it lists no MoCap for v3/s1, v3/s2, v3/s5, v3/s6, which do have MoCap). Use metadata/recordings.csv and each recording's alignment_metadata.json instead.

Changelog

v1.2.1 (October 2026) — documentation and metadata corrections; sensor data, annotations and videos unchanged.
- README, DATASHEET and croissant.json corrected: final title, venue, authors, contact and maintenance plan; EMG described as the raw signed signal on the right forearm; MoCap, EyeTrack and IMU columns described as released; video file names and frame rates; 5-fold cross-validation protocol (the split column is legacy).
- metadata/recordings.csv: v19/s5 no longer lists scene_video (only its gaze-overlay video exists). metadata/modality_coverage.xlsx: v5/s8 and v18/s2 marked as without first-person video. Coverage figures regenerated (Scene Cam 92.1%).
- The eyetrack and imu Dataset Viewer configs were removed (their column sets differ between recordings); load these files with pandas.
v1.2 (October 2026) — S9 motion-primitive labels.
- Added human-reviewed primitive-level labels for 30 S9 recordings (454 segments, 16 composite primitives): annotations/v*/s9.json and annotations_flat/segments_s9.csv (see S9 motion-primitive labels). S1–S8 annotations and all sensor data are unchanged.
v1.1 (October 2026) — annotation refresh.
- All action segments replaced by the human-reviewed and corrected annotations described in Annotation provenance: 7,789 → 7,356 segments, still 282 annotated recordings (v16/s5 is now annotated; v18/s2 no longer is).
- Fixes to the previous labels: malformed timestamps (e.g. 01:00:02 written for 00:01:02), abbreviated hand labels with missing fields in 44 segments, and non-English text in 36 segment descriptions.
- v32/s1 and v30/s6 trimmed (see Known issues); metadata/recordings.csv and metadata/modality_coverage.xlsx updated accordingly.
- annotations_flat/segments.csv, metadata/annotations/segment_counts.{csv,xlsx} and the README figures regenerated from the released annotations.
- Description of the cross-modal time alignment updated: streams are aligned from each device's recorded start timestamp (see Synchronization).
v1.0 (May 2026) — initial release.

Repository layout

PULSE/
├── README.md            ← this file
├── DATASHEET.md         ← Gebru-style data card
├── LICENSE              ← CC BY-NC 4.0 + use restrictions
├── CODE_LICENSE         ← MIT, applies to the companion code repository PULSE-code
├── croissant.json       ← Croissant machine-readable card with RAI fields
├── .gitattributes       ← LFS rules (do not delete before re-upload)
├── assets/              ← static figures used by this README
├── annotations/
│   └── v*/s*.json                    ← per-recording nested JSON (canonical)
├── annotations_flat/
│   ├── segments.csv                  ← flat per-segment table (Dataset Viewer subset `action_segments`)
│   └── segments_s9.csv               ← S9 motion-primitive labels (subset `s9_primitive_segments`)
├── data/
│   └── v*/s*/
│       ├── aligned_<modality>_100hz.csv  (5 core + 2 Myo aux)
│       ├── alignment_metadata.json
│       ├── raw/      → opt-in Qualisys skeleton TSV (~32 MB / file, ~8 GB total)
│       └── videos/   → opt-in scene_cam.mp4 + eye_tracking_overlay.mp4 (~200 MB / pair, ~63 GB total)
├── metadata/
│   ├── recordings.csv                ← full 337-row recording manifest
│   ├── modality_coverage.xlsx
│   ├── batch_alignment_summary.json  ← legacy batch log (see Known issues)
│   └── annotations/
│       └── segment_counts.{csv,xlsx}

Ethics, privacy, and intended use

IRB / ethics committee approval obtained from the South China University of Technology (approval no. S-2025-026-01, 4 March 2025). Data were recorded at South China University of Technology, Guangzhou, China.
Written informed consent + separate media-release agreement for the scene-cam and eye-tracking videos; released videos contain no audio track, and the face of any incidentally captured laboratory member is pixelated.
The release does not include participant demographic attributes (sex, age, height, weight) collected for screening.
Use restrictions: see LICENSE. Surveillance, employee monitoring, and biometric re-identification applications without freely given consent are explicitly disallowed.
Downstream use of biometric-class modalities (EMG, eye tracking, pressure) for inference of sensitive attributes (fatigue, stress, medical conditions) on new subjects requires the deploying institution's own ethics review.
Withdrawal and takedown requests: contact the corresponding author (below).

Contact and maintenance

Authors: Wolin Liang, Yang Gao, Peiyu Yan, Yuexiang Hu, Yingjing Xiao, Xiongfeng Ying, Yingnian Guo, Cheng Zhang, Zhanpeng Jin — School of Future Technology, South China University of Technology (Yingjing Xiao: East China Normal University).
Corresponding author and dataset contact: Yang Gao, gaoyang2025@scut.edu.cn
Maintainer: Wolin Liang, 202610196903@mail.scut.edu.cn
Maintenance plan: the dataset is hosted on Hugging Face under the permanent DOI 10.57967/hf/11314, and the authors will host and maintain it for at least five years after publication. Errata are listed under Known issues and fixes are released as new versions with an entry in the Changelog. Questions and bug reports can also be posted in the Community (Discussions) tab of this page.

Citation

@inproceedings{liang2026pulse,
  title     = {PULSE: A Synchronized Five-Modality Dataset for Sensorimotor Coordination in Long-Horizon Daily Activities},
  author    = {Wolin Liang and Yang Gao and Peiyu Yan and Yuexiang Hu and Yingjing Xiao and
               Xiongfeng Ying and Yingnian Guo and Cheng Zhang and Zhanpeng Jin},
  booktitle = {NeurIPS 2026 Evaluations and Datasets Track},
  year      = {2026},
  url       = {https://openreview.net/forum?id=AoU576r9O2}
}

Dataset DOI: 10.57967/hf/11314

Downloads last month: 1,468

Models trained or fine-tuned on velvet-pine-22/PULSE

velvet-pine-22/PULSE-code

Updated 4 days ago