This presentation is accessible only to the Illinois State University community.
- Off-Campus ISU Users: To download this item, click the "Off-Campus Download" button below. You will be prompted to log in with your ISU ULID and password.
Publication Date
2023
Document Type
Poster
Degree Type
Undergraduate
Mentor
Mehdi Karimi
Abstract
● A project in Deep Reinforcement Learning (RL), a popular branch in Machine Learning (ML) with several modern applications in robotics.
● We use deep RL and the Unity platform to program and design a physically simulated spider. We use Unity’s Machine Learning Agents (ML-Agents) package and PyTorch to train the spider to complete certain tasks.
● Completed training tasks: ○ Follow an object ○ Turn to face a specific directions
● Unity ML-Agents lets us define the RL agents, the environment, action, and reward functions. It also allows us to tweak hyper-parameters which dictate neural network structures and training properties.
● The project additionally includes studying theoretical aspects of RL to better understand the training process.
Recommended Citation
Beyer, Daniel and Wittrock, Joseph, "Training a Physically Simulated Virtual Spider" (2023). University Research Symposium. 419.
https://ir.library.illinoisstate.edu/rsp_urs/419