Datasets:

arthur801031
/

mail

batch_size int64 8 8	best_epoch int64 0 0 ⌀	best_valid_loss int64 0 0 ⌀	beta1 float64 0.9 0.9	cached_states_path stringclasses 2 values	collect_data_delta_move_max float64 0.4 1	collect_data_delta_move_min float64 0.15 1	copy_teach sequencelengths 2 2	dataf stringclasses 2 values	down_sample_scale int64 3 3	dt float64 0.01 0.01	edge_model_path null	env_name stringclasses 2 values	exp_name stringclasses 1 value	fix_collision_edge bool 1 class	fixed_lr bool 1 class	full_dyn_path null	full_lr float64 0 0	global_size int64 128 128	imit_w int64 5 5	imit_w_lat int64 1 1	load_optim bool 1 class	log_dir stringclasses 2 values	lr float64 0 0	n_epoch int64 1k 1k	n_his int64 5 5	n_rollout int64 2k 2k	neighbor_radius float64 0.05 0.05	nstep_eval_rollout int64 20 20	num_variations int64 100 100	num_workers int64 10 10	partial_dyn_path null	partial_observable bool 1 class	particle_radius float64 0.01 0.01	pred_time_interval int64 5 5	proc_layer int64 10 10	relation_dim int64 7 7	reward_w float64 100k 100k	save_model_interval int64 5 5	seed int64 100 100	shape_state_dim int64 14 14	state_dim int64 18 18	time_step int64 100 100	train_mode stringclasses 1 value	train_valid_ratio float64 0.9 0.9	tune_teach bool 1 class	use_collision_as_mesh_edge bool 1 class	use_mesh_edge bool 1 class	use_rest_distance bool 1 class	use_wandb bool 1 class	voxel_size float64 0.02 0.02	vsbl_lr float64 0 0
8	0	0	0.9	ours_clothfold_n100.pkl	0.4	0.15	[ "encoder", "decoder" ]	./data/ours_clothfold_vcd_11	3	0.01	null	ClothFold	test	false	false	null	0.0001	128	5	1	false	data/Clothfold_GNS_12.07.10.59_11	0.0001	1,000	5	2,000	0.045	20	100	10	null	true	0.00625	5	10	7	100,000	5	100	14	18	100	vsbl	0.9	false	false	true	true	false	0.0216	0.0001
8	null	null	0.9	ours_clothfold_n100.pkl	0.4	0.15	[ "encoder", "decoder" ]	./data/ours_clothfold_vcd_11	3	0.01	null	ClothFold	test	false	false	null	0.0001	128	5	1	false	data/Clothfold_GNS_12.07.10.59_11	0.0001	1,000	5	2,000	0.045	20	100	10	null	true	0.00625	5	10	7	100,000	5	100	14	18	100	vsbl	0.9	false	false	true	true	false	0.0216	0.0001
8	0	0	0.9	ours_drycloth_n100.pkl	1	1	[ "encoder", "decoder" ]	./data/drycloth_vcd_12.06.19.09_11	3	0.01	null	DryCloth	test	false	false	null	0.0001	128	5	1	false	data/DryCloth_GNS_12.11.09.57_11	0.0001	1,000	5	2,000	0.045	20	100	10	null	true	0.00625	5	10	7	100,000	5	100	14	18	100	vsbl	0.9	false	false	true	true	false	0.0216	0.0001

Learning Robot Manipulation from Cross-Morphology Demonstration (CoRL 2023)

[Project website] [arXiv PDF]

Datasets for MAIL.

Authors: Gautam Salhotra*, I-Chun Arthur Liu*, Gaurav S. Sukhatme (* denotes equal contribution)

Some Learning from Demonstrations (LfD) methods handle small mismatches in the action spaces of the teacher and student. Here we address the case where the teacher’s morphology is substantially different from that of the student. Our framework, Morphological Adaptation in Imitation Learning (MAIL), bridges this gap allowing us to train an agent from demonstrations by other agents with significantly different morphologies. MAIL learns from suboptimal demonstrations, so long as they provide some guidance towards a desired solution. We demonstrate MAIL on manipulation tasks with rigid and deformable objects including 3D cloth manipulation interacting with rigid obstacles. We train a visual control policy for a robot with one end-effector using demonstrations from a simulated agent with two end-effectors. MAIL shows up to 24% improvement in a normalized performance metric over LfD and non-LfD baselines. It is deployed to a real Franka Panda robot, handles multiple variations in properties for objects (size, rotation, translation), and cloth-specific properties (color, thickness, size, material).

Downloads last month: 10

Size of the auto-converted Parquet files:

59.4 kB

Number of rows:

Paper for arthur801031/mail

Learning Robot Manipulation from Cross-Morphology Demonstration

Paper • 2304.03833 • Published Apr 7, 2023