Voinea, Andrei (2022) Scalable Distributed Learning with PPG-IMPALA for Physics-based Musculoskeletal Models. Bachelor's Thesis, Artificial Intelligence.
|
Text
AndreiVoinea_s3754243_Bachelor_Thesis.pdf Download (1MB) | Preview |
|
Text
toestemming.pdf Restricted to Registered users only Download (102kB) |
Abstract
The use of Deep Reinforcement Learning (DRL) algorithms has surged in previous years due to their ability to adapt to a variety of tasks. However, training these algorithms sometimes requires long periods of time, and the resource efficiency can be low. To this end, the current study evaluates the use of an off-policy DRL algorithm, which can efficiently use a large number of resources, during a bipedal motion control task. The proposed algorithm is used to generate a gait pattern during the simulation of a physics-based musculoskeletal model of a healthy subject. As the goal was to achieve normal speed level-ground walking, the training process was assisted using imitation data provided by a public dataset where participants performed such movements. Although the trained policy could not generate a stable gait pattern, the results indicate that the proposed architecture can increase the speed of training, when compared with an on-policy architecture.
Item Type: | Thesis (Bachelor's Thesis) |
---|---|
Supervisor name: | Carloni, R. |
Degree programme: | Artificial Intelligence |
Thesis type: | Bachelor's Thesis |
Language: | English |
Date Deposited: | 16 Aug 2022 07:45 |
Last Modified: | 29 Aug 2022 10:39 |
URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28394 |
Actions (login required)
View Item |