Robots with Privacy Stipulations

Abstract

In late July last year, it came to light that iRobot Corp. intended to sell the maps that modern Roomba vacuum cleaning robots build to help them navigate. This caused a public furor among consumers. This situation and several others (e.g., nuclear inspection, use of untrusted cloud computing infrastructure) suggest that we might be interested in limiting what information a robot might divulge. How should we think about robotic privacy? In this talk I’ll describe a line of research that is concerned with this question, starting by showing that cryptography doesn’t solve the problem. I’ll begin by examining a privacy-preserving tracking task, then look at how one might think about estimators that are constrained to ensure they never know too much. Finally, I’ll talk about planning subject to information disclosure constraints and introduce a useful structure that we call a “plan closure.”