AMR24-FP6-DCOL

This repo contains the code for a differentiable collision detection system, using several shapes as convex primitives.

Table of Contents

• Introduction
• Installation
• Usage
• Results
• Systems
• Acknowledgements

Introduction

This project aims to develop a differentiable collision detection system, re-implementing in Python this Julia library

Installation

To install and set up this project locally, follow these steps:

1. Clone the repository:

   git clone https://github.com/DIAG-Robotics-Lab/AMR24-FP6-DCOL.git

2. Navigate to the project directory:

   cd AMR24-FP6-DCOL

3. (optional) Create a virtual environment

   python -m venv venv
   source venv/bin/activate

   This will keep things cleaner

4. Install dependencies:

   pip install -r requirements.txt

Usage

After installation, you can use the project as follows:

1. Run the application: Run the script speciying the system you want to simulate

   python main.py --system <system_name>

Systems

The available systems in this repo are

• Piano Mover: a piano must maneuver around a 90-degree turn in a hallway. The walls are 1 meter apart, and the “piano” (a line segment) is 2.6 meters long, making the path around the corner nontrivial.
• Quadrotor: a classic 6-DOF quadrotor model traverses a cluttered hallway with 12 objects in it.
• Cone Through a Wall: a cone must be routed through a square hole in a wall.

Results

Piano Mover

Running

python main.py --system piano_mover

the result is the following:

Quadrotor

Running

python main.py --system quadrotor

the result is the following:

Cone Through Wall

Running

python main.py --system coneThroughWall

the result is the following:

Acknowledgements

• Differentiable Collision Detection for a Set of Convex Primitives
• ALTRO Algorithm
• iLQR Tutorial