wp99cp / navigation_event_prediction

Predicting Events based on Offline-Data Collected with the ANYmal Research Robot

This is part of my course project for the course Foundations of Reinforcement Learning at ETH.

How to run the Code?

1. Clone the repository
2. Install the requirements
3. Save the data inside the data folder and adjust the paths in main.py in order to load the correct .bagfiles.
4. Run the main.py file

This will pre-process the data, then saves it as a .pkl file (which can be loaded later) and then trains the model. Pre-processing the data is a time-consuming process.

What is the project about?

Abstract

We investigate using offline reinforcement learning to enhance the navigation capabilities of legged robots using previously collected real-world data. Our approach utilizes a comprehensive dataset from past ANYmal missions to develop a helper component for navigation that predicts the likelihood of navigational events based on sensor inputs.

This project is part of the course project for the course Foundations of Reinforcement Learning at ETH.