NOTE: this code is based heavily on the Ravens code base from Google and retains the same license.

Integration of Embodied AI for Tangle-Free Operation in Robot Vacuums

The main contributions of our paper are as follows:

• We proposed a novel method to integrate the segmentation and rearrangement of deformable linear objects to identify and move cables.
• We designed rigid reward metrics with comprehensive constraints to meaningfully train our agent in a manner that closely mimics real-world scenarios so that it is able to solve the entanglement problem in reality.
• The results validated the coherence of each aforementioned metric between our theoretical framework, indicating the effectiveness of our method.

| Installation | Environments and Tasks | Quick Start |

---

Installation

This is how to get the code running on a local machine. First, get conda on the machine if it isn't there already:

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh

Then, create a new Python 3.7 conda environment (e.g., named "py3-defs") and activate it:

conda create -n py3-defs python=3.7
conda activate py3-defs

Then install:

./install_python_ubuntu.sh

Environments and Tasks

In adapting the original repository to align with our specific task requirements, we made crucial modifications to defs_cable.py. These changes focused on adjusting both the parameters representing the radius of the beads and the number of parts.

Furthermore, we undertook modifications in environment.py to facilitate the extraction of the robot's position to compute the reward function. Our task-specific reward function comprises three decisions, integrated into the Task super-class.

Quick Start

We provide the following example for users to quickly implement our model.

# Run with default configurations
python main.py