2/n
We use Gaussian Splats for safe robot navigation: Splats model the scene, and their Gaussians act as repulsive barriers via control barrier functions to prevent collisions given requested inputs from the human teleoperator.

2/n We efficiently build a high-quality Gaussian Splat by capturing initial views, generating a Splat, and using pose uncertainty in color and depth to guide the next best real-world view for scene reconstruction.

@peasant98.bsky.social

JPARSE for handling robot manipulator singularities!
jparse-manip.github.io
#AcademicSky #EDUSky #Robots

youtu.be/ZMrNGQht9eg?...

youtu.be/ZMrNGQht9eg?...

(6/n) To learn more, read our paper and use our open-source code base, check out our interactive paper website: jparse-manip.github.io

(5/n) Our method J-PARSE stands to improve robotic teleoperation control, improve embodiment in imitation learning, and allow robot upper limbs to move in a more 'natural' way like human counterparts.

(4/n) This approach is motivated by the anthropomorphic motion of athletes such as baseball players who extend their hand towards their target for catching a ball, and they only consider the motion of their hand in this fast motion control.

(3/n) Fewer methods still take you into singular poses and require very specific knowledge of the robot and its particular singular poses, as well as an accompanying strategy to exit these poses.

(2/n) Robot manipulator singularities have been a challenge for roboticists for decades. Over the years, there have been many approaches to primarily avoid them and, in a few cases (such as damped least squares), approach them, albeit sacrificing pose accuracy at non-singular poses.

youtu.be/ZMrNGQht9eg?...

(6/n) To learn more, read our paper and use our open-source code base, check out our interactive paper website: jparse-manip.github.io and our paper video: youtu.be/ZMrNGQht9eg?...

(5/n) Our method J-PARSE stands to improve robotic teleoperation control, improve embodiment in imitation learning, and allow robot upper limbs to move in a more 'natural' way like human counterparts.

(4/n) This approach is motivated by the anthropomorphic motion of athletes such as baseball players who extend their hand towards their target for catching a ball, and they only consider the motion of their hand in this fast motion control.

(3/n) Fewer methods still take you into singular poses and require very specific knowledge of the robot and its particular singular poses, as well as an accompanying strategy to exit these poses.

(2/n) Robot manipulator singularities have been a challenge for roboticists for decades. Over the years, there have been many approaches to primarily avoid them and, in a few cases (such as damped least squares), approach them, albeit sacrificing pose accuracy at non-singular poses.

Latex (overleaf) is the way! Handles formatting and reference styles. For ref.bib:

\documentclass{article}
\usepackage[utf8]{inputenc}
\begin{document}

This is a citation \cite{example}.

\bibliographystyle{apa} % Switch this to "vancouver" or another style
\bibliography{ref}

\end{document}

Our work appeared in #CORL2024