Robust autonomy:

Overview

When autonomous systems such as self-driving cars and robotic manipulators are deployed in real-world environments, it is of the utmost importance to consider---and ideally to guarantee---safe runtime operation. Since these systems often operate in highly uncertain and dynamic environments, it is crucial for them to model and quantify environmental uncertainty, understand its impact on system dynamics, predict the motion of other agents, and make safe, risk-aware decisions. Safety and robustness have been studied extensively from a theoretical perspective, and there are several prominent success stories in application, e.g. in aviation. However, techniques with strong theoretical safety properties have yet to penetrate many new and exciting robotic application areas, such as autonomous driving, in which uncertainty in environmental perception and prediction overwhelm traditional safety analysis.

Topics and Objectives

This workshop aims to:

1. raise open questions on safety issues when robots operate autonomously in uncertain, real-world environments
2. discuss meaningful theoretical relaxations of strict safety guarantees which could be more easily used in practice
3. encourage conversation between perception and control communities on handling uncertainty from the sensors, down to actuation, and
4. provide a forum for discussion among researchers, industry, and regulators as to the core challenges, promising solution strategies, fundamental limitations, and regulatory realities involved in deploying safety-critical systems

Areas of interest:

Modeling uncertainty, safe motion planning, collision avoidance, decision-making in dynamic environments, intent prediction, safe exploration, safety and risk analysis, etc.

Techniques include:

Optimal control, Robust control, Probability theory, Bayesian inference, POMDPs, etc.