Deep Reinforcement Learning for Physics-based Musculoskeletal Model of a Transfemoral Amputee with a Prosthesis Walking on Uneven Terrain

Boer, Sarah de (2021) Deep Reinforcement Learning for Physics-based Musculoskeletal Model of a Transfemoral Amputee with a Prosthesis Walking on Uneven Terrain. Bachelor's Thesis, Artificial Intelligence.

Preview	Text Thesis_SarahdeBoer.pdf Download (2MB) | Preview
	Text toestemming.pdf Restricted to Registered users only Download (104kB)

Abstract

This paper focuses on deep reinforcement learning for physics-based musculoskeletal model of a transfemoral amputee with a prosthesis walking on uneven terrain. A multilayer perceptron is used with as learning algorithm proximal policy optimization in combination with imitation learning. The imitation data, of a healthy subject walking on flat surface, is gathered from an open dataset and preprocessed to fit the model. The physics-based musculoskeletal model of a transfemoral amputee subject has a prosthesis on the left leg. This prosthesis contains two actuators, one at the knee joint and one at the ankle joint. The model is simulated using the opensource simulation software OpenSim. Two versions of the model are used, one with Coordinate actuators (OpenSim4.1) and one with Activation Coordinate actuators (OpenSim4.2). Using the deep reinforcement learning algorithm, the model is taught how to perform a humanlike gait on an uneven ground. The robustness of the learning algorithm is being tested using the musculoskeletal model and the muscle activations are observed when walking on uneven terrain instead of level-ground walking. No gait pattern could be observed, hence, no conclusions could be made regarding muscle activation when walking on uneven terrain.

Item Type:	Thesis (Bachelor's Thesis)
Supervisor name:	Carloni, R. and Raveendranathan, V.
Degree programme:	Artificial Intelligence
Thesis type:	Bachelor's Thesis
Language:	English
Date Deposited:	29 Jul 2021 08:37
Last Modified:	10 Jan 2024 13:15
URI:	https://fse.studenttheses.ub.rug.nl/id/eprint/25498

Actions (login required)

View Item