pickplace_sac_youbot.gin

# Training & playing the pick and place task with agent youbot_noplugin
#
# DISPLAY=null python -m alf.bin.train --root_dir=~/tmp/pickplace --gin_file=pickplace_sac_youbot.gin
# python -m alf.bin.play --root_dir=~/tmp/pickplace --gin_file=pickplace_sac_youbot.gin

include 'common_nonimage_sac.gin'
import social_bot

# environment config
create_environment.env_name="SocialBot-PlayGround-v0"
create_environment.num_parallel_environments=12
create_environment.env_load_fn=@suite_socialbot.load
# suite_socialbot.load.gym_env_wrappers=(@FrameStack,)

PlayGround.agent_type='youbot_noplugin'
PlayGround.tasks=[@PickAndPlace]
PlayGround.max_steps=150
suite_socialbot.load.max_episode_steps=152
PlayGround.step_time=0.05
PickAndPlace.reward_shaping=True
PlayGround.action_wrapper=@YoubotActionWrapper

# algorithm config
actor/Adam.learning_rate=1e-3
critic/Adam.learning_rate=1e-3
alpha/Adam.learning_rate=1e-3
OneStepTDLoss.gamma=0.98

# training config
TrainerConfig.initial_collect_steps=2000
TrainerConfig.mini_batch_length=2
TrainerConfig.unroll_length=1
TrainerConfig.mini_batch_size=512
TrainerConfig.summary_interval=200
TrainerConfig.eval_interval=2000
TrainerConfig.checkpoint_interval=2000
TrainerConfig.num_updates_per_train_step=1

TFUniformReplayBuffer.max_length=100000