Gymnasium api - #1

Open

korneelf1 wants to merge 76 commits into

tudelft:mainfrom

korneelf1:Gymnasium-API

Commits on Jun 10, 2024

Custom Gym Env, Crazyflie simulator, sb3/tianshou progress, custom kernels
Korneel Van den Berghe
committed
147b61a

Commits on Jun 17, 2024

gymnasium API simulator
Korneel Van den Berghe
committed
3169338
Improved docstrings
Korneel Van den Berghe
committed
2ae4e38
support for actin history, numpy version of deque, slows down slightly
Korneel Van den Berghe
committed
c2056c6
adapt reward function to target stabilizing controller rather than position controller
Korneel Van den Berghe
committed
a700384
Warn if action not in actionspace, might cause NaN values
Korneel Van den Berghe
committed
2837bcf

Commits on Jun 18, 2024

enable position term in reward, test and train scripts
Korneel Van den Berghe
committed
909e07b

Commits on Jun 19, 2024

updates on n_episodes (for testing)
Korneel Van den Berghe
committed
d4d56c3
trained policy
korneelf1
committed
06fd50b
imu model, changed reward structure
Korneel Van den Berghe
committed
0895da6
changed reward str
Korneel Van den Berghe
committed
70fdab0
Merge branch 'gymnasium_env' of https://github.com/korneelf1/fastPyDroneSim_gym into gymnasium_env
Korneel Van den Berghe
committed
e4c9bfd
matplotlib rendering
Korneel Van den Berghe
committed
3949c2d
mac policy
korneelf1
committed
cf6bd60
fix test envs + stabalization reward
Korneel Van den Berghe
committed
92e2f00
adapt train script
Korneel Van den Berghe
committed
e52ecd2

Commits on Jun 24, 2024

improved collector, progressed gpu kernels
korneelf1
committed
ddd1219
stabalization ctrl
korneelf1
committed
cdd7482
Prioritized buffer, plot origin point for rendering
korneelf1
committed
61da397
Cuda support for kernels
korneelf1
committed
6ce17a5

Commits on Jun 25, 2024

Recurrent training
korneelf1
committed
e15b93d
Trained policies
korneelf1
committed
4be9604
position control
korneelf1
committed
6b8fb11

Commits on Jun 26, 2024

Commits on Jun 27, 2024

restore training characteristics to last_working branch?
korneelf1
committed
3d6577b

Commits on Jul 25, 2024

Spiking Actors, Updated Evotorch
korneelf1
committed
fa52928
wandb logging of evotorch
korneelf1
committed
e7854c9

Commits on Jul 26, 2024

Commits on Jul 29, 2024

spiking actor implemented
korneelf1
committed
9863f5a

Commits on Jul 30, 2024

SMLP is now nn.Module, evotorch adapted to support SNN
korneelf1
committed
0da6c0d
save the evolutionary model
korneelf1
committed
43a7b19

Commits on Jul 31, 2024

debugging by adding option to specify task
korneelf1
committed
c30c968
swap around n_states
korneelf1
committed
18794db
more info on whats happening in terminal
korneelf1
committed
b63284b

Commits on Aug 1, 2024

debugging....
korneelf1
committed
e7863a8

Commits on Aug 2, 2024

Fix speed issue, add masked networks
korneelf1
committed
a87fd1b

Commits on Aug 4, 2024

!!! thrust constant and torque constant were swapped around (k is thrust constant which is tiny)
korneelf1
committed
e1ff0ab

Commits on Aug 5, 2024

Trying out og drone
korneelf1
committed
ba4f150
less strict termination conditions to speed up training
korneelf1
committed
3fb2767
I wasnt logging the args...
korneelf1
committed
ec19a26
Cumulative reward if evaluation env
korneelf1
committed
54595d2

Commits on Aug 8, 2024

Commits on Aug 9, 2024

implement standard collector compatibility
korneelf1
committed
9e523a5
remove device arg in collector, change hyperparameters to ensure at least 3M interactions (see learning to fly)
korneelf1
committed
9d9e9b0
batch size of 1 in PPO apperently causes it to crash :)
korneelf1
committed
048aed6
batch size of 1 does NOT work!
korneelf1
committed
247c160
ANN evo learning
korneelf1
committed
48a7226
Fix logging, fix PPO (action bounded)
korneelf1
committed
6c2b343

Commits on Aug 12, 2024

Increase learning rate
korneelf1
committed
374c6e1
Tianshou to evotorch wrapper
korneelf1
committed
04e5e89

Commits on Aug 13, 2024

Normalize motor speeds
korneelf1
committed
0939233
Remove test statements for reward printing
korneelf1
committed
3481caa
Cant remove psets for now
korneelf1
committed
1aafac8
N=1 different return in step
korneelf1
committed
32a1376
reward function from git reward_squared_fast_learning
korneelf1
committed
795a669
dont normalize, 1e-4 learning rate with no scheduling
korneelf1
committed
7ad36bd
allow tianshou policy with evotorch
korneelf1
committed
03926f8

Commits on Aug 14, 2024

Improved reset mechanism, initially completely random, but might reset to impossible states
korneelf1
committed
b6ae688
Allwo multidim quaternion init
korneelf1
committed
9842ea6
allow multiple lr schedulers
korneelf1
committed
d4fb96e
step size of 1k for lr schedulers
korneelf1
committed
9c58498
format gym env
korneelf1
committed
a17f5c5
CLEANUP: clean up gym_sim file, helpers file
korneelf1
committed
def3b11
Something of a test
korneelf1
committed
9549371
Cleaned up file structure
korneelf1
committed
ac0b222
Syncing with OG repo
korneelf1
committed
dece0bf
test dones and single steps
korneelf1
committed
67d0e6a

Commits on Aug 15, 2024

Fix of unittests
korneelf1
committed
7174ac3