Talking Robotics


Organizers: Patrícia Alves-Oliveira,
Silvia Tulli, Miguel Vasco
contact us: talkingrobotics at gmail dot com — support us: buymeacoffe

twitter youtube


Daniel Rakita is a Ph.D. student of computer science at the University of Wisconsin-Madison advised by Michael Gleicher and Bilge Mutlu. He received a Bachelors of Music Performance from the Indiana University Jacobs School of Music in 2012. His research lies at the intersection of motion planning, trajectory optimization, shared-control, and human-robot interaction, and commonly involves creating motion optimization approaches that allow robot manipulators to move smoothly and accurately in real-time.

Speaker Links: Website - Google Scholar


In order to complete tasks, robots need to consistently calculate joint-space trajectories such that their end-effectors pass through the correct position, with the correct orientation, at the correct time. In many cases, these trajectories must not only be accurate, i.e., avoiding large end-effector translation or orientation errors; and feasible, i.e., avoiding self-collisions, joint-space discontinuities, or kinematic singularities; but they must also be calculated quickly to afford robots the ability to act and react in uncertain or dynamic environments. In this talk, I will overview technical methods we have developed that attempt to achieve such feasible, accurate, and time-sensitive robot-arm motions. In particular, I will detail our inverse kinematics solver called RelaxedIK that utilizes both non-linear optimization and machine learning to achieve a smooth, feasible, and accurate end-effector to joint-space mapping on-the-fly. I will highlight numerous ways we have applied our technical methods to real-world-inspired problems, such as mapping human-arm-motion to robot-arm-motion in real-time to afford effective shared-control interfaces and automatically moving a camera-in-hand robot in a remote setting to optimize a viewpoint for a teleoperator. I will conclude with a preview of our ongoing and future work in this space.

Papers covered during the talk