pedestrian_prediction

Realtime, confidence-varying trajectory prediction for "Probabilistically Safe Robot Planning with Confidence-Based Human Predictions". RSS '18

Installation

git clone https://github.com/sirspinach/pedestrian_prediction.git
cd pedestrian_prediction
pip install -e .

Module contents

• pp.mdp: Data structures and value iteration algorithms for two types of GridWorlds -- a standard GridWorld with only lateral and diagonal movement, and a GridWorldExpanded in which actions operate in "gridless" continuous space but are snapped to the nearest grid cell.
• pp.inference: Algorithms for inferring destinations, occupancies, and states given a GridWorld, a list of destinations, and the pedestrian's trajectory so far.
• pp.util: Utilities for parsing and trajectory manipulation.
• pp.plots: Plot simulated trajectories, overlayed with heatmaps of predicted states. Produce some test plots by executing python -m pp.plot. (Requires plotly)