umautobots / osp

Code for Off The Beaten Sidewalk paper (https://arxiv.org/abs/2006.00962)

Off The Beaten Sidewalk: Pedestrian Prediction In Shared Spaces For Autonomous Vehicles

by Cyrus Anderson at UM FCAV

Introduction

This paper presents the method Off the Sidewalk Predictions (OSP) to predict pedestrians' trajectories in scenes where sidewalks and other traffic devices may not be present (such as shared spaces).

arxiv: https://arxiv.org/abs/2006.00962

Predict Trajectories

Predictions with pre-trained models can be made by running

python driver_low_mem.py

Model Fitting

Model parameters can be estimated from data by running

python ss_model/fit_model_driver.py

File Structure

The structure at SAMPLE_DATASETS_ROOT:

sample_data
   | tt_format
      | 10hz
          | dut

Additional datasets can be resampled and formatted with the tools in utils/dataset_conversion.py. The pedestrian datasets used in the paper are from:

• DUT
• InD Dataset (apply there)

Dependencies

• numpy
• scipy
• pandas