-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathDeveloper Notes.txt
57 lines (39 loc) · 3.04 KB
/
Developer Notes.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
cd catkin_workspace/
source devel/setup.bash
rosdep install openai_ros
cd
cd workspace/ga-drl-aubo/
python3 ga.py
roslaunch aubo_i5_moveit_config demo.launch
python2.7 moveit_motion_control.py
python3 -m train --env=AuboReach-v3 --logdir=$HOME/openaiGA --n_epochs=20 --num_cpu=4
python3 -m train --env=AuboReach-v1 --logdir=$HOME/openaiGA --n_epochs=100 --num_cpu=6 --polyak_value=0.184 --gamma_value=0.88 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.055 --noise_epsilon=0.774
python3 -m train --env=AuboReach-v2 --logdir=$HOME/openaiGA --n_epochs=100 --num_cpu=6 --polyak_value=0.184 --gamma_value=0.88 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.055 --noise_epsilon=0.774
python3 -m train --env=AuboReach-v2 --logdir=$HOME/openaiGA --n_epochs=100 --num_cpu=6 --polyak_value=0.555 --gamma_value=0.22 --q_learning=0.444 --pi_learning=0.333 --random_epsilon=0.777 --noise_epsilon=0.111
# Execution 10 found ga parameters 10.23.2021
python3 -m train --env=AuboReach-v2 --logdir=$HOME/openaiGA-test --n_epochs=10 --num_cpu=6 --polyak_value=0.004 --gamma_value=0.193 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.814 --noise_epsilon=0.94
python3 -m train --env=AuboReach-v2 --logdir=$HOME/openaiGA-test --n_epochs=10 --num_cpu=6 --polyak_value=0.897 --gamma_value=1.048 --q_learning=0.372 --pi_learning=0.705 --random_epsilon=0.988 --noise_epsilon=0.135
python3 -m train --env=AuboReach-v3 --logdir=$HOME/openaiGA --n_epochs=20 --num_cpu=4 --polyak_value=0.924 --gamma_value=0.949 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.584 --noise_epsilon=0.232
python3 -m train --env=AuboReach-v2 --logdir=$HOME/openaiGA-test --n_epochs=30 --num_cpu=6 --polyak_value=0.209 --gamma_value=0.287 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.405 --noise_epsilon=0.247
# ga found params
python3 -m train --env=AuboReach-v5 --logdir=$HOME/openaiGA --n_epochs=30 --num_cpu=1 --polyak_value=0.924 --gamma_value=0.988 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.912 --noise_epsilon=0.748
# original
python3 -m train --env=AuboReach-v5 --logdir=$HOME/openaiGA-test --n_epochs=30 --num_cpu=1 --polyak_value=0.95 --gamma_value=0.98 --q_learning=0.001 --pi_learning=0.001 --random_epsilon=0.3 --noise_epsilon=0.2
python3 plot.py $HOME/openaiGA
python3 play-v2.py $HOME/openaiGA/policy_best.pkl
Parameters changed for Testing:
max_episode_steps #100 aubo_reach2.py
success rate threshold #0.85 train.py, aubo_reach2.py
self.T in rollouts.py -> causing one cycle in epoch to run forever #100
n_cycles #50 config.py, DEFAULT_ENV_PARAMS
n_batches #40 config.py
epochs_default #50 ga.py
n_test_rollouts #50 config.py
population_size #30 ga.py
ga.run #30 ga.py
ga run (30) -> epoch (50) -> cycle (50) -> episode (100) -> self.T in rollout.py (500) -> batch (40)
# mobile manipulator code
https://github.com/robowork/autonomous_mobile_manipulation
# forward kinematics to convert joint states to end effector pose
https://answers.ros.org/question/327873/find-end-effector-pose-from-joint-state/
https://github.com/orocos/orocos_kinematics_dynamics