Deep Reinforcement Learning for Gait generation of a Musculoskeletal Model in a Physics-Based Simulation on Even and Uneven Terrain.

Luinge, Rutger (2022) Deep Reinforcement Learning for Gait generation of a Musculoskeletal Model in a Physics-Based Simulation on Even and Uneven Terrain. Bachelor's Thesis, Artificial Intelligence.

Preview	Text DRL_walking_simulation_thesis_R_Luinge_v2 (5).pdf Download (1MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (125kB)

Abstract

This research focused on the use of a deep reinforcement learning algorithm to train a musculoskeletal model to walk on even and uneven terrain. To achieve this goal a learning agent was used, which we were able to teach based on the agent his past experiences, the objective of this agent was to generate a healthy gait pattern. This goal was achieved by using proximal policy optimization in combination with imitation learning. The data used in this research consists of the joint positions and velocities of a healthy walking subject during a walking trial, taken from an open source data set. This data was scaled to the model, which consisted of 22 muscles and 14 degrees of freedom. The different muscles in the musculoskeletal model could be actuated using the open source software OpenSim4.3. By using this software and the described learning architecture, the model was taught to walk on even terrain. The model/policy was subsequently transferred to uneven terrain to test the stability and capabilities of the architecture, and to simulate a more realistic scenario. As a result the model was able to walk and show a healthy gait cycle. When the model got transferred to the uneven terrain, the agent was still able to perform gait cycles but on the other hand the gait showed less stability throug not staying upright after the first two gaits. At the cost of knee flexion the algorithm was able to make the agent walk on the uneven terrain. The left side of the musculoskeletal model se

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Carloni, R.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	20 Sep 2022 11:57
Last Modified:	20 Sep 2022 11:57
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/28733

Actions (login required)

View Item