CLoSD / evaluation /closd /config.yaml
guytevet's picture
added generated data for evaluation from /home/dcor/sigalraab/train_outputs/rl/from_guy/PrefixCompHor_DiffusionSteps_10_PredLen_40_new/in_paper/model000200000.DiMP_cfg5_crop16_len316_expT2mFinetune_planh40_switchProb0.0_recover2
de7106b
!!python/object/new:easydict.EasyDict
dictitems:
add_proj: false
debug: false
device: cuda
device_id: 0
device_type: cuda
dimp_eval: false
env: &id030 !!python/object/new:easydict.EasyDict
dictitems:
asset_type: mesh
auto_pmcp: false
auto_pmcp_soft: true
aux_coefficient: 0.3
bench_asset_density: 1000.0
contact_bodies: &id001
- L_Knee
- L_Ankle
- L_Toe
- R_Knee
- R_Ankle
- R_Toe
- Torso
- Chest
- L_Thorax
- L_Shoulder
- L_Elbow
- L_Wrist
- L_Hand
- R_Thorax
- R_Shoulder
- R_Elbow
- R_Wrist
- R_Hand
controlFrequencyInv: 2
control_mode: isaac_pd
cycle_motion: false
cycle_motion_xp: false
debug: &id002 !!python/object/new:easydict.EasyDict
dictitems:
no_aux_obs: false
no_terrain: false
state:
no_aux_obs: false
no_terrain: false
dimp: &id003 !!python/object/new:easydict.EasyDict
dictitems:
cfg_param: 5
context_switch_prob: 0.0
debug_hml: false
dump_mdm: false
handle_text_distrib: true
limit_context: null
max_saved_hml_episodes: 50000
model_path: /home/dcor/sigalraab/train_outputs/rl/from_guy/PrefixCompHor_DiffusionSteps_10_PredLen_40_new/model000200000.pt
open_loop_hook: false
planning_horizon: 40
planning_horizon_multiplyer: 1
prefix_crop_size: 16
prompt: A person is walking.
recon_frame_start: 0
recon_frame_stop: -1
recon_param: 10000.0
recon_step_start: -1
recon_step_stop: 2
save_hml_episodes: true
save_hml_episodes_dir: ''
save_name_suffix: recover2
traj_guidance: false
state:
cfg_param: 5
context_switch_prob: 0.0
debug_hml: false
dump_mdm: false
handle_text_distrib: true
limit_context: null
max_saved_hml_episodes: 50000
model_path: /home/dcor/sigalraab/train_outputs/rl/from_guy/PrefixCompHor_DiffusionSteps_10_PredLen_40_new/model000200000.pt
open_loop_hook: false
planning_horizon: 40
planning_horizon_multiplyer: 1
prefix_crop_size: 16
prompt: A person is walking.
recon_frame_start: 0
recon_frame_stop: -1
recon_param: 10000.0
recon_step_start: -1
recon_step_stop: 2
save_hml_episodes: true
save_hml_episodes_dir: ''
save_name_suffix: recover2
traj_guidance: false
disable_adaptnet: true
done_dist: 0.3
enableEarlyTermination: true
enableTaskObs: true
enable_debug_vis: false
enable_transitions: true
env_spacing: 5
episode_length: 300
fut_tracks: false
hard_negative: false
hybridInitProb: 0.5
imitator: HumanoidIm
is_flag_run: false
key_bodies: &id004
- R_Ankle
- L_Ankle
- R_Wrist
- L_Wrist
kp_scale: 1
local_root_obs: true
min_length: 5
models: output/HumanoidIm/PhcSinglePrimForever/Humanoid.pth
motion_file: sample_data/amass_isaac_standing_upright_slim.pkl
notes: ''
numAMPObsSteps: 10
numActions: 69
numObservations: 574
numTrajSamples: 3
num_envs: 4096
obs_v: 7
plane: &id005 !!python/object/new:easydict.EasyDict
dictitems:
dynamicFriction: 1.0
restitution: 0.0
staticFriction: 1.0
state:
dynamicFriction: 1.0
restitution: 0.0
staticFriction: 1.0
power_reward: true
power_scale: 1.0
project_name: PHC
reachBodyName: Pelvis
reset_bodies: &id006
- Pelvis
- L_Hip
- L_Knee
- R_Hip
- R_Knee
- Torso
- Spine
- Chest
- Neck
- Head
- L_Thorax
- L_Shoulder
- L_Elbow
- L_Wrist
- L_Hand
- R_Thorax
- R_Shoulder
- R_Elbow
- R_Wrist
- R_Hand
root_height_obs: true
seed: -1
self_obs_v: 1
shape_resampling_interval: 100000000000000000
show_markers: true
stateInit: Random
strikeBodyNames: &id007
- L_Hand
strike_asset_density: 200.0
support_phc_markers: true
tarChangeStepsMax: 100
tarChangeStepsMin: 50
tarDistMax: 1.0
tarHeightMax: 0.95
tarHeightMin: 0.9
tarSpeed: 1.0
task: HumanoidAdaptNetT2M
terminationDistance: 0.25
terminationHeight: 0.15
terrain: &id008 !!python/object/new:easydict.EasyDict
dictitems:
height_safety_gap: 0.0
state:
height_safety_gap: 0.0
trajSampleTimestepInv: 3
state:
asset_type: mesh
auto_pmcp: false
auto_pmcp_soft: true
aux_coefficient: 0.3
bench_asset_density: 1000.0
contact_bodies: *id001
controlFrequencyInv: 2
control_mode: isaac_pd
cycle_motion: false
cycle_motion_xp: false
debug: *id002
dimp: *id003
disable_adaptnet: true
done_dist: 0.3
enableEarlyTermination: true
enableTaskObs: true
enable_debug_vis: false
enable_transitions: true
env_spacing: 5
episode_length: 300
fut_tracks: false
hard_negative: false
hybridInitProb: 0.5
imitator: HumanoidIm
is_flag_run: false
key_bodies: *id004
kp_scale: 1
local_root_obs: true
min_length: 5
models: output/HumanoidIm/PhcSinglePrimForever/Humanoid.pth
motion_file: sample_data/amass_isaac_standing_upright_slim.pkl
notes: ''
numAMPObsSteps: 10
numActions: 69
numObservations: 574
numTrajSamples: 3
num_envs: 4096
obs_v: 7
plane: *id005
power_reward: true
power_scale: 1.0
project_name: PHC
reachBodyName: Pelvis
reset_bodies: *id006
root_height_obs: true
seed: -1
self_obs_v: 1
shape_resampling_interval: 100000000000000000
show_markers: true
stateInit: Random
strikeBodyNames: *id007
strike_asset_density: 200.0
support_phc_markers: true
tarChangeStepsMax: 100
tarChangeStepsMin: 50
tarDistMax: 1.0
tarHeightMax: 0.95
tarHeightMin: 0.9
tarSpeed: 1.0
task: HumanoidAdaptNetT2M
terminationDistance: 0.25
terminationHeight: 0.15
terrain: *id008
trajSampleTimestepInv: 3
epoch: 61500
exp_name: T2mFinetune
follow: true
has_eval: true
headless: true
horovod: false
im_eval: false
learning: &id031 !!python/object/new:easydict.EasyDict
dictitems:
params: &id026 !!python/object/new:easydict.EasyDict
dictitems:
algo: &id022 !!python/object/new:easydict.EasyDict
dictitems:
name: im_amp
state:
name: im_amp
config: &id023 !!python/object/new:easydict.EasyDict
dictitems:
amp_batch_size: 512
amp_minibatch_size: 4096
amp_obs_demo_buffer_size: 200000
amp_replay_buffer_size: 200000
amp_replay_keep_prob: 0.01
bounds_loss_coef: 10
clip_value: false
critic_coef: 5
dimp_eval: false
disc_coef: 5
disc_grad_penalty: 5
disc_logit_reg: 0.01
disc_reward_scale: 2
disc_reward_w: 0.5
disc_weight_decay: 0.0001
e_clip: 0.2
entropy_coef: 0.0
env_name: rlgpu
gamma: 0.99
grad_norm: 50.0
horizon_length: 32
learning_rate: 2.0e-05
lr_schedule: constant
max_epochs: 10000000
mini_epochs: 6
minibatch_size: 16384
mixed_precision: false
multi_gpu: false
name: Humanoid
network_path: output/HumanoidIm/T2mFinetune
normalize_advantage: true
normalize_amp_input: true
normalize_input: true
normalize_value: true
num_actors: 4096
player: &id009 !!python/object/new:easydict.EasyDict
dictitems:
games_num: 50000000
print_stats: false
state:
games_num: 50000000
print_stats: false
ppo: true
print_stats: false
reward_shaper: &id010 !!python/object/new:easydict.EasyDict
dictitems:
scale_value: 1
state:
scale_value: 1
save_best_after: 100
save_frequency: 500
save_intermediate: true
score_to_win: 20000
task_reward_w: 0.5
tau: 0.95
train_dir: output/HumanoidIm/T2mFinetune
truncate_grads: true
state:
amp_batch_size: 512
amp_minibatch_size: 4096
amp_obs_demo_buffer_size: 200000
amp_replay_buffer_size: 200000
amp_replay_keep_prob: 0.01
bounds_loss_coef: 10
clip_value: false
critic_coef: 5
dimp_eval: false
disc_coef: 5
disc_grad_penalty: 5
disc_logit_reg: 0.01
disc_reward_scale: 2
disc_reward_w: 0.5
disc_weight_decay: 0.0001
e_clip: 0.2
entropy_coef: 0.0
env_name: rlgpu
gamma: 0.99
grad_norm: 50.0
horizon_length: 32
learning_rate: 2.0e-05
lr_schedule: constant
max_epochs: 10000000
mini_epochs: 6
minibatch_size: 16384
mixed_precision: false
multi_gpu: false
name: Humanoid
network_path: output/HumanoidIm/T2mFinetune
normalize_advantage: true
normalize_amp_input: true
normalize_input: true
normalize_value: true
num_actors: 4096
player: *id009
ppo: true
print_stats: false
reward_shaper: *id010
save_best_after: 100
save_frequency: 500
save_intermediate: true
score_to_win: 20000
task_reward_w: 0.5
tau: 0.95
train_dir: output/HumanoidIm/T2mFinetune
truncate_grads: true
load_checkpoint: true
load_path: output/HumanoidIm/T2mFinetune/Humanoid_00061500.pth
model: &id024 !!python/object/new:easydict.EasyDict
dictitems:
name: amp
state:
name: amp
network: &id025 !!python/object/new:easydict.EasyDict
dictitems:
adapt_actor: true
adapt_critic: false
disc: &id019 !!python/object/new:easydict.EasyDict
dictitems:
activation: relu
initializer: &id011 !!python/object/new:easydict.EasyDict
dictitems:
name: default
state:
name: default
units: &id012
- 1024
- 512
state:
activation: relu
initializer: *id011
units: *id012
mlp: &id020 !!python/object/new:easydict.EasyDict
dictitems:
activation: silu
d2rl: false
initializer: &id013 !!python/object/new:easydict.EasyDict
dictitems:
name: default
state:
name: default
regularizer: &id014 !!python/object/new:easydict.EasyDict
dictitems:
name: None
state:
name: None
split_after_layer: 0
units: &id015
- 2048
- 1536
- 1024
- 1024
- 512
- 512
state:
activation: silu
d2rl: false
initializer: *id013
regularizer: *id014
split_after_layer: 0
units: *id015
name: amp_adapt
separate: true
space: &id021 !!python/object/new:easydict.EasyDict
dictitems:
continuous: &id018 !!python/object/new:easydict.EasyDict
dictitems:
fixed_sigma: true
learn_sigma: false
mu_activation: None
mu_init: &id016 !!python/object/new:easydict.EasyDict
dictitems:
name: default
state:
name: default
sigma_activation: None
sigma_init: &id017 !!python/object/new:easydict.EasyDict
dictitems:
name: const_initializer
val: -2.9
state:
name: const_initializer
val: -2.9
state:
fixed_sigma: true
learn_sigma: false
mu_activation: None
mu_init: *id016
sigma_activation: None
sigma_init: *id017
state:
continuous: *id018
state:
adapt_actor: true
adapt_critic: false
disc: *id019
mlp: *id020
name: amp_adapt
separate: true
space: *id021
seed: 0
state:
algo: *id022
config: *id023
load_checkpoint: true
load_path: output/HumanoidIm/T2mFinetune/Humanoid_00061500.pth
model: *id024
network: *id025
seed: 0
state:
params: *id026
metadata: false
no_log: false
no_virtual_display: true
notes: Default Notes
num_threads: 64
output_path: output/HumanoidIm/T2mFinetune
play: true
project_name: PHC
real_traj: false
render_o3d: false
resume_str: null
rl_device: cuda:0
robot: &id032 !!python/object/new:easydict.EasyDict
dictitems:
asset: &id027 !!python/object/new:easydict.EasyDict
dictitems:
assetFileName: mjcf/smpl_humanoid.xml
assetRoot: /
state:
assetFileName: mjcf/smpl_humanoid.xml
assetRoot: /
bias_offset: false
big_ankle: true
box_body: true
freeze_hand: false
freeze_toe: false
has_dof_subset: true
has_jt_limit: false
has_mesh: false
has_self_collision: true
has_shape_obs: false
has_shape_obs_disc: false
has_shape_variation: false
has_smpl_pd_offset: false
has_upright_start: true
humanoid_type: smpl
masterfoot: false
motion_sym_loss: false
real_weight: true
real_weight_porpotion_boxes: true
real_weight_porpotion_capsules: true
remove_toe: false
sym_loss_coef: 1
state:
asset: *id027
bias_offset: false
big_ankle: true
box_body: true
freeze_hand: false
freeze_toe: false
has_dof_subset: true
has_jt_limit: false
has_mesh: false
has_self_collision: true
has_shape_obs: false
has_shape_obs_disc: false
has_shape_variation: false
has_smpl_pd_offset: false
has_upright_start: true
humanoid_type: smpl
masterfoot: false
motion_sym_loss: false
real_weight: true
real_weight_porpotion_boxes: true
real_weight_porpotion_capsules: true
remove_toe: false
sym_loss_coef: 1
seed: -1
server_mode: false
sim: &id033 !!python/object/new:easydict.EasyDict
dictitems:
flex: &id028 !!python/object/new:easydict.EasyDict
dictitems:
num_inner_iterations: 10
warm_start: 0.25
state:
num_inner_iterations: 10
warm_start: 0.25
graphics_device_id: 0
physx: &id029 !!python/object/new:easydict.EasyDict
dictitems:
bounce_threshold_velocity: 0.2
contact_offset: 0.02
default_buffer_size_multiplier: 10.0
max_depenetration_velocity: 10.0
num_position_iterations: 4
num_threads: 4
num_velocity_iterations: 0
rest_offset: 0.0
solver_type: 1
state:
bounce_threshold_velocity: 0.2
contact_offset: 0.02
default_buffer_size_multiplier: 10.0
max_depenetration_velocity: 10.0
num_position_iterations: 4
num_threads: 4
num_velocity_iterations: 0
rest_offset: 0.0
solver_type: 1
pipeline: gpu
sim_device: cuda:0
slices: 0
subscenes: 0
substeps: 2
use_flex: false
state:
flex: *id028
graphics_device_id: 0
physx: *id029
pipeline: gpu
sim_device: cuda:0
slices: 0
subscenes: 0
substeps: 2
use_flex: false
test: true
torch_deterministic: false
train: false
state:
add_proj: false
debug: false
device: cuda
device_id: 0
device_type: cuda
dimp_eval: false
env: *id030
epoch: 61500
exp_name: T2mFinetune
follow: true
has_eval: true
headless: true
horovod: false
im_eval: false
learning: *id031
metadata: false
no_log: false
no_virtual_display: true
notes: Default Notes
num_threads: 64
output_path: output/HumanoidIm/T2mFinetune
play: true
project_name: PHC
real_traj: false
render_o3d: false
resume_str: null
rl_device: cuda:0
robot: *id032
seed: -1
server_mode: false
sim: *id033
test: true
torch_deterministic: false
train: false