TA_ALP-GMM_SAC_spider_s1

ClementRomac/TA_ALP-GMM_SAC_spider_s1

TA_ALP-GMM_SAC_spider_s1 is a reinforcement learning model from ClementRomac. Use it for the reinforcement learning task on the model card, and read the license before you ship it in a product.

You can find more info about TeachMyAgent here.

Downloads · 30 days

Access

Public

Updated Aug 11, 2022

Repo size

2.3 MB

Likes

Public

Hugging Face

Repo makeup

Click a slice to open those files.

.bin2.3 MB · 99%

At a glance

Task: Reinforcement Learning
Access: Public
Created: Aug 11, 2022
Updated: Aug 11, 2022
SHA: 0d02edfc

Model card

From the Hugging Face model README

Deep RL Agent Playing TeachMyAgent's parkour.

You can find more info about TeachMyAgent here.

Results of our benchmark can be found in our paper.

You can test this policy here.

This policy was not part of TeachMyAgent's benchmark

Results

Percentage of mastered tasks (i.e. reward >= 230) after 20 millions steps on the Parkour track.

Results shown are averages over 16 seeds along with the standard deviation for each morphology as well as the aggregation of the 48 seeds in the Overall column.

We highlight the best results in bold.

Algorithm	BipedalWalker	Fish	Climber	Overall
Random	27.25 (± 10.7)	23.6 (± 21.3)	0.0 (± 0.0)	16.9 (± 18.3)
ADR	14.7 (± 19.4)	5.3 (± 20.6)	0.0 (± 0.0)	6.7 (± 17.4)
ALP-GMM	42.7 (± 11.2)	36.1 (± 28.5)	0.4 (± 1.2)	26.4 (± 25.7)
Covar-GMM	35.7 (± 15.9)	29.9 (± 27.9)	0.5 (± 1.9)	22.1 (± 24.2)
GoalGAN	25.4 (± 24.7)	34.7 ± 37.0)	0.8 (± 2.7)	20.3 (± 29.5)
RIAC	31.2 (± 8.2)	37.4 (± 25.4)	0.4 (± 1.4)	23.0 (± 22.4)
SPDL	30.6 (± 22.8)	9.0 (± 24.2)	1.0 (± 3.4)	13.5 (± 23.0)
Setter-Solver	28.75 (± 20.7)	5.1 (± 7.6)	0.0 (± 0.0)	11.3 (± 17.9)

Hyperparameters

{'student': 'SAC'
'environment': 'parkour'
'training_steps': 20000000
'n_evaluation_tasks': 100
'teacher': 'ALP-GMM'
'morphology': 'spider'}

Task: Reinforcement Learning
Created: Aug 11, 2022
Updated: Aug 11, 2022