Lange, Carl (2022) Towards Smooth Policies in Deep Reinforcement Learning for Musculoskeletal Simulations of Human Walking. Bachelor's Thesis, Artificial Intelligence.
|
Text
Thesis_Carl_Lange_s4007433.pdf Download (5MB) | Preview |
|
![[img]](https://fse.studenttheses.ub.rug.nl/style/images/fileicons/text.png) |
Text
toestemming.pdf Restricted to Registered users only Download (125kB) |
Abstract
Recent advances in the musculoskeletal modeling of human walking demonstrated the ability of Proximal Policy Optimization combined with imitation learning to train a simulated musculoskeletal model for human-like walking. However, the resulting gait featured erratically changing muscle activations, which imposes a hurdle for applying this technique to actuated prostheses. Therefore, the main goal of this paper was to build on the previous work and apply regularization techniques in order to learn a smoother policy with less erratic activations. The results demonstrate that L2 layer regularization, changed neuron activation function, and reward shaping effectively decreased muscle erraticness and resulting torques. However, not all simulation runs achieved a stable gait pattern. Additionally, the learned gaits that were stable showed large discrepancies in joint angles when compared to human data. In summary, the results suggest that the applied regularization techniques were successful at decreasing erraticness but there exists a trade-off between regularization and performance.
| Item Type: | Thesis (Bachelor's Thesis) |
|---|---|
| Supervisor name: | Carloni, R. |
| Degree programme: | Artificial Intelligence |
| Thesis type: | Bachelor's Thesis |
| Language: | English |
| Date Deposited: | 02 Sep 2022 14:47 |
| Last Modified: | 02 Sep 2022 14:47 |
| URI: | https://fse.studenttheses.ub.rug.nl/id/eprint/28596 |
Actions (login required)
 |
View Item |