forked from tangger/lerobot
c623824139
- Added Nan detection mechanisms in the actor, learner and gym_manipulator for the case where we encounter nans in the loop. - changed the non-blocking in the `.to(device)` functions to only work for the case of cuda because they were causing nans when running the policy on mps - Added some joint clipping and limits in the env, robot and policy configs. TODO clean this part and make the limits in one config file only. Co-authored-by: Adil Zouitine <adilzouitinegm@gmail.com>
29 lines
694 B
YAML
29 lines
694 B
YAML
# @package _global_
|
|
|
|
fps: 30
|
|
|
|
env:
|
|
name: real_world
|
|
task: null
|
|
state_dim: 6
|
|
action_dim: 6
|
|
fps: ${fps}
|
|
device: mps
|
|
|
|
wrapper:
|
|
crop_params_dict:
|
|
observation.images.laptop: [58, 89, 357, 455]
|
|
observation.images.phone: [3, 4, 471, 633]
|
|
resize_size: [128, 128]
|
|
control_time_s: 20
|
|
reset_follower_pos: true
|
|
use_relative_joint_positions: true
|
|
reset_time_s: 5
|
|
display_cameras: false
|
|
delta_action: 0.1
|
|
joint_masking_action_space: [1, 1, 1, 1, 0, 0] # disable wrist and gripper
|
|
|
|
reward_classifier:
|
|
pretrained_path: outputs/classifier/checkpoints/best/pretrained_model
|
|
config_path: lerobot/configs/policy/hilserl_classifier.yaml
|
|
|