Events Calendar

CS Colloquium: Alberto Rodriguez: Contacting the World with Mechanical and Data-Driven Intelligence

Speaker: Alberto Rodriguez, Carnegie Mellon

Talk Title: Contacting the World with Mechanical and Data-Driven Intelligence

Series: CS Colloquium

Abstract: In the next 10 to 20 years, society will look towards robotics to solve some of its biggest challenges: from improving the self-sufficiency of an aging population, to enabling more efficient and intelligent manufacturing processes; from assisting in dangerous environmental cleanup operations, to providing immediate support in search and rescue emergencies. Reliable physical interaction is central to all these challenges, and robots must master it to become part of the solution.

End effectors such as robotic hands play a privileged role in the manipulation chain. They contact the world, and both their designs and their actions can contribute to more intelligent and reliable physical interaction. In my research I explore both. The central idea is to combine the simplicity and reliability of end effectors and control strategies designed to exhibit mechanical intelligence with the realism of data-driven models to give robots the necessary skills to expect, understand, and control contact.

In the first part of the talk I will discuss the role of the design of a mechanism in producing intelligent behavior. I will show how mechanical attributes such as shape, actuation, or compliance, can be instrumental in the design of effectors that are simpler, cheaper, lighter, and more reliable, and how exploring their design tradeoffs has the potential to impact a very broad set of applications, from automating the design of specialized grippers to the design of feet and locomotion gaits that take into consideration the statistics of the terrain.

In the second part of the talk I will address the problem of getting robots to control physical interaction through their actions. Contact leaves a trace of sensor readings that a skilled manipulator should be able to understand to direct its actions. I will show how we can build accurate probabilistic data-driven models for perception, planning, prediction, and failure detection, to direct and monitor the execution of manipulation tasks, with example applications to general-purpose in-hand manipulation and automated assembly.

Biography: Alberto Rodriguez is a Ph.D. candidate at the Robotics Institute at Carnegie Mellon University. His main research interests are in robotic manipulation, including mechanical design, data-driven manipulation, grasping, caging, and automated assembly. His long-term research goal is to provide robots with enough sensing, reasoning and acting capabilities to reliably manipulate the environment. Alberto Rodriguez received the degrees of Mathematics ('05) and Telecommunication Engineering ('06 with honors) from the Universitat Politecnica de Catalunya (UPC) in Barcelona, Spain. He is the recipient of "La Caixa" and "Caja Madrid" fellowships for graduate studies in the US, and the recipient of the Best Student Paper Award at the conference Robotics: Science and Systems 2011.

Host: Fei Sha

Location: Seaver Science Library (SSL) - 150

Audiences: Everyone Is Invited

Contact: Assistant to CS chair