Real-Time Certified Probabilistic Pedestrian Forecasting
Henry O. Jacobs · Owen Hughes · Matthew Johnson-Roberson · Ram Vasudevan
TL;DR
Pedestrian motion is forecast probabilistically using weighted sums of learned differential equations, running in real time and beating prior methods over long horizons.
BRIEF
Self-driving cars and other robots need to anticipate where pedestrians and cyclists will go, and how uncertain that guess is, before they decide how to move. This work learns a set of simple motion models, written as ordinary differential equations, from past trajectories recorded in one fixed scene. Predictions come from a weighted sum of these models, giving a probability over future locations. The method runs in parallel, works in real time even in plain Python, and is more accurate than existing approaches over long horizons.
Abstract
The success of autonomous systems will depend upon their ability to safely navigate human-centric environments. This motivates the need for a real-time, probabilistic forecasting algorithm for pedestrians, cyclists, and other agents since these predictions will form a necessary step in assessing the risk of any action. This letter presents a novel approach to probabilistic forecasting for pedestrians based on weighted sums of ordinary differential equations that are learned from historical trajectory information within a fixed scene. The resulting algorithm is embarrassingly parallel and is able to work at real-time speeds using a naive Python implementation. The quality of predicted locations of agents generated by the proposed algorithm is validated on a variety of examples and is considerably higher than existing state of the art approaches over long time horizons.