We present a Reverse Reinforcement Learning (Reverse RL) approach for representing retrospective knowledge .

We present a Reverse Reinforcement Learning (Reverse RL) approach for representing retrospective knowledge. General Value Functions (GVFs) have enjoyed great success in representing predictive knowledge, i.e., answering questions about possible future outcomes such as "how much fuel will be consumed in expectation if we drive from A to B?". GVFs, however, cannot answer questions like "how much fuel do we expect a car to have given it is at B at time $t$?". To answer this question, we need to know when that car had a full tank and how that car came to B. Since such questions emphasize the influence of possible past events on the present, we refer to their answers as retrospective knowledge. In this paper, we show how to represent retrospective knowledge with Reverse GVFs, which are trained via Reverse RL. We demonstrate empirically the utility of Reverse GVFs in both representation learning and anomaly detection.

Microdrones has announced that, as part of an ongoing global growth initiative, it has acquired Aircam UAV Technology, a 64 employee Chinese company that provides UAV (unmanned aerial vehicle) technologies and services. Aircam has developed a large Chinese and Southeast Asian customer base with a focus on surveying & mapping, utilities, and oil & gas industries. Aircam will be fully integrated with the Microdrones business, brand and leadership team. The Aircam brand and corporate identity will change to Microdrones, and all aspects of the business will be directed by the Microdrones global leadership team. Microdrones and Aircam have a long history of working together.

Last month, we wrote about autonomous quadrotors from the University of Pennsylvania that use just a VGA camera and an IMU to navigate together in swarms. Without relying on external localization or GPS, quadrotors like these have much more potential to be real-world useful, since they can operate without expensive and complex infrastructure, even indoors. One potential application for drones like these is disaster operations, but honestly, that's just what everyone says when you ask them how their mobile robot could potentially be useful. What's much more interesting to us are commercial applications, and with drones, that inevitably means talking about delivery. There are a lot of reasons why we're skeptical about most commercial delivery drones, but that doesn't meant that the idea of using drones to move things from place to place isn't a good one. Vijay Kumar's lab at UPenn has been working on using their GPS-independent quadrotors for transporting payloads, and they're doing it collaboratively--the idea is that objects that are too large or heavy for one quadrotor to move can instead be moved by multiple quadrotors working together, and ultimately, they could be the best way to move items around a warehouse.