Machine learning for human robot collaboration
With the advent of technology, autonomous robotic agents have attained an increasingly important role in various industries such as manufacturing, transportation or even agriculture. Despite the numerous benefits they offer, these agents still present the challenge of adapting to small changes in th...
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Final Year Project |
| Language: | English |
| Published: |
Nanyang Technological University
2022
|
| Subjects: | |
| Online Access: | https://hdl.handle.net/10356/158190 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Institution: | Nanyang Technological University |
| Language: | English |
| Summary: | With the advent of technology, autonomous robotic agents have attained an increasingly important role in various industries such as manufacturing, transportation or even agriculture. Despite the numerous benefits they offer, these agents still present the challenge of adapting to small changes in their tasks. An additional challenge is encountered in the effort and expertise required to teach an agent a new skill. This process can be simplified through Imitation Learning, which aims to teach an agent new skills by exposing it to expert demonstrations. This project explores the use of Generative Adversarial Imitation Learning to train a robotic simulation on the OpenAI Gym framework. Further, this project examines the effectiveness of pre-training and Spectral Normalization on the performance of the trained agent and the speed and stability of the training process. Through extensive experimentation, this project determines that using Proximal Policy Optimization instead of Trust-Region Policy Optimization can enhance agent performance. Additionally, this project shows that pre-training can accelerate learning speed, and spectral normalization can improve training stability. |
|---|