README.md

ORCA Sim

Generate synthetic pedestrian datasets using ORCA (optimal reciprocal collision avoidance).

Introduction

This directory contains the code used to generate our synthetic dataset for training and evaluating TRACE. It makes use of Python bindings for the original ORCA simulator.

Installation

Some extra set up is required to install the ORCA simulator.

First, install a few extra requirements

pip install -r requirements.txt

Then, clone the ORCA sub-repo using:

cd Python-RVO2
git submodule init
git submodule update

And install it:

python setup.py build
python setup.py install
cd ..

Data Generation

Here are the commands used to generate the two subsets of our synthetic ORCA dataset:

# ORCA Maps data generation
python gen_dataset.py --num_scenes 1000 --scene_len 10 --max_agents 10 --max_obstacles 20 --viz --out datagen_out/orca_sim/maps --resample_goal

# ORCA no maps data generation
python gen_dataset.py --num_scenes 1000 --scene_len 10 --max_agents 20 --max_obstacles 0 --viz --out datagen_out/orca_sim/no_maps --resample_goal

You can alternatively use the --viz_vid flag to render videos of the simulations.