Robotics Reaps Rewards at ICRA: NVIDIA’s Dieter Fox Named RAS Pioneer


Hundreds of researchers from all the plot in which thru the globe shall be gathering — unbiased about — next week for the IEEE World Convention on Robotics and Automation.

As a flagship conference on all issues robotics, ICRA has develop to be a familiar forum since its inception in 1984. This yr, NVIDIA’s Dieter Fox will receive the RAS Pioneer Award, given by the IEEE Robotics and Automation Society.

Fox is the firm’s senior director of robotics evaluate and head of the NVIDIA Robotics Analysis Lab, in Seattle, moreover to a professor at the College of Washington Paul G. Allen College of Computer Science & Engineering and head of the UW Robotics and Remark Estimation Lab. At the NVIDIA lab, Fox leads over 20 researchers and interns, fostering collaboration with the neighboring UW.

He’s receiving the RAS Pioneer Award “for pioneering contributions to probabilistic say estimation, RGB-D realizing, machine finding out in robotics, and bridging academic and industrial robotics evaluate.”

“Being identified with this award by my evaluate colleagues and the IEEE society is an unprecedented honor,” Fox stated. “I’m very grateful for the amazing collaborators and college students I had the likelihood to work with for the length of my career. I also delight in that IEEE sees the importance of connecting academic and industrial evaluate — I relate that bridging these areas lets in us to make faster development on the complications we genuinely care about.”

Fox would possibly possibly even give a chat at the conference, the set an whole of 19 papers that investigate a differ of topics in robotics shall be presented by researchers from NVIDIA Analysis.

Right here’s a preview of some of the showcase-stopping NVIDIA evaluate papers that had been licensed at ICRA:

Robotics Work a Finalist for Simplest Paper Awards

6-DOF Greedy for Plan-Driven Object Manipulation in Clutter” is a finalist for every and each the Simplest Paper Award in Robotic Manipulation and the Simplest Pupil Paper.

The paper delves into the no longer easy robotics remark of greedy in cluttered environments, which is a necessity in most valid-world scenes, stated Adithya Murali, one of many lead researchers and a graduate scholar at the Robotics Institute at Carnegie Mellon College. Considerable fresh evaluate considers handiest planar greedy, in which a robot grasps from the head down as a substitute of transferring in extra dimensions.

Arsalan Mousavian, one other lead researcher on the paper and a senior evaluate scientist at the NVIDIA Robotics Analysis Lab, outlined that they performed this evaluate in simulation. “We weren’t sure by any bodily robot, which is time-drinking and very expensive,” he stated.

Mousavian and his colleagues professional their algorithms on NVIDIA V100 Tensor Core GPUs, after which examined on NVIDIA TITAN GPUs. For this specific paper, the coaching records consisted of simulating 750,000 robot object interactions in lower than half of a day, and the models had been professional in per week. Once professional, the robot used to be in a location to robustly manipulate objects within the valid world.

Replanning for Success

NVIDIA Analysis also thought to be how robots would possibly possibly per chance notion to carry out a wide differ of initiatives in no longer easy environments, akin to greedy an object that isn’t visible, in a paper called “On-line Replanning in Perception Arena for Partially Observable Job and Ride Complications.”

The absolute best plot makes a differ of initiatives that that chances are you’ll relate. Caelan Garrett, graduate scholar at MIT and a lead researcher on the paper, outlined, “Our work is reasonably usual in that we address initiatives that involve no longer handiest picking and placing issues within the environment, however also pouring issues, cooking, attempting to starting up doorways and drawers.”

Garrett and his colleagues created an starting up-provide algorithm, SS-Replan, that lets within the robot to incorporate observations when making selections, which it’ll regulate per contemporary observations it makes whereas attempting to carry out its purpose.

They examined their work in NVIDIA Isaac Sim, a simulation surroundings feeble to make, test and decide into consideration digital robots, and on an genuine robot.

DexPilot: A Teleoperated Robotic Hand-Arm Machine

In one other paper, NVIDIA researchers confronted the difficulty that fresh robotics algorithms don’t yet allow for a robot to total accurate initiatives akin to pulling a tea discover out of a drawer, laying aside a greenback invoice from a wallet or unscrewing the lid off a jar autonomously.

In “DexPilot: Depth-Based Teleoperation of Dexterous Robotic Hand-Arm Machine,” NVIDIA researchers fresh a draw in which a human can remotely operate a robotic draw. DexPilot observes the human hand the usage of cameras, after which uses neural networks to relay the motion to the robotic hand.

Whereas other systems require expensive instruments akin to motion-defend systems, gloves and headsets, DexPilot archives teleoperation thru a combination of deep finding out and optimization.

It took 15 hours to put collectively on a single GPU after we collected the records, basically basically basically based on NVIDIA researchers Ankur Handa and Karl Van Wyk, two of the authors of the paper. They and their colleagues feeble the NVIDIA TITAN GPU for his or her evaluate.

Learn all about these papers and more at ICRA 2020.

The NVIDIA evaluate team has more than 200 scientists all the plot in which thru the globe, centered on areas akin to AI, computer vision, self-driving vehicles, robotics and graphics. Learn more at

Leave a comment

Your email address will not be published.