Learning to Forecast Aleatoric and Epistemic Uncertainties over Long Horizon Trajectories

TL;DR

This paper introduces a deep generative world model that forecasts long-term trajectories of autonomous agents while quantifying both aleatoric and epistemic uncertainties, enhancing safety and communication of agent capabilities.

Contribution

The work presents a novel deep generative model that jointly estimates aleatoric and epistemic uncertainties over long horizon trajectories in autonomous systems.

Findings

Calibrated uncertainty estimates over full trajectories

Effective quantification of aleatoric uncertainty

Incorporation of epistemic uncertainty during learning

Abstract

Giving autonomous agents the ability to forecast their own outcomes and uncertainty will allow them to communicate their competencies and be used more safely. We accomplish this by using a learned world model of the agent system to forecast full agent trajectories over long time horizons. Real world systems involve significant sources of both aleatoric and epistemic uncertainty that compound and interact over time in the trajectory forecasts. We develop a deep generative world model that quantifies aleatoric uncertainty while incorporating the effects of epistemic uncertainty during the learning process. We show on two reinforcement learning problems that our uncertainty model produces calibrated outcome uncertainty estimates over the full trajectory horizon.