- 325
- 219 519
GRASP Lab
Registrace 21. 11. 2007
The General Robotics, Automation, Sensing and Perception (GRASP) Laboratory is an interdisciplinary academic and research center within the School of Engineering and Applied Sciences at the University of Pennsylvania. Founded in 1979, the GRASP Lab is a premier robotics incubator that fosters collaboration between students, research staff and faculty focusing on fundamental research in vision, perception, control systems, automation, and machine learning.
Spring 2024 GRASP SFI Harish Ravichandar, Georgia Institute of Technology
“New Wine in an Old Bottle: A Structured Approach to Democratize Robot Learning”
ABSTRACT
Decades of rigorous research in dynamical systems and control helped us integrate robots into a wide variety of domains, ranging from factory floors to the moon. Today, it would appear that deep learning has taken over the torch and will bring robots to our homes, freeing us all from banal chores. In this utopian vision, learning-based approaches tend to replace analytical methods. Moving away from handcrafted bespoke solutions to generalist robots that can operate in unstructured environments. But one can instead view learning-based and analytical approaches as two ends of a broad spectrum, with one end optimizing for reliability (at the cost of human effort) and the other for emergent intelligence (at the cost of data and computation). In this talk, I will argue why it is better for robots to be in the middle of this broad spectrum. Using manipulation as a case study, I will discuss how our lab combines ideas from dynamical systems and machine learning to overcome three often-overlooked issues with contemporary methods: i) high barrier to entry due to demands for expensive computational resources and annotated data, ii) inability to handle new tasks without relying on significant user expertise (e.g., for reward or controller design, hyperparameter tuning, data collection and curation), and iii) unreliable behaviors due to inscrutable and unpredictable learned policies. Addressing these issues will enable robot learning to escape the confines of well-resourced research labs and positively impact the larger society.
PRESENTER
Harish Ravichandar is an Assistant Professor in the School of Interactive Computing and a core faculty member of the Institute for Robotics and Intelligent Machines (IRIM) at Georgia Institute of Technology. He directs the Structured Techniques for Algorithmic Robotics (STAR) Lab, where his team leverages ideas from dynamical systems and control to design structured robot learning algorithms that improve reliability, efficiency, and self-sufficiency of robots operating in unstructured human environments. His research is motivated by and pertinent to diverse applications, ranging from dexterous manipulation to multi-agent coordination. His work has been recognized by IEEE MRS Best Paper Award, ASME DSCC Best Student Paper Award, IEEE CSS Video Contest Award, UTC Institute for Advanced System Engineering Graduate Fellowship, and Georgia Tech’s College of Computing Outstanding Post-Doctoral Research and Outstanding Research Scientist Awards.
ABSTRACT
Decades of rigorous research in dynamical systems and control helped us integrate robots into a wide variety of domains, ranging from factory floors to the moon. Today, it would appear that deep learning has taken over the torch and will bring robots to our homes, freeing us all from banal chores. In this utopian vision, learning-based approaches tend to replace analytical methods. Moving away from handcrafted bespoke solutions to generalist robots that can operate in unstructured environments. But one can instead view learning-based and analytical approaches as two ends of a broad spectrum, with one end optimizing for reliability (at the cost of human effort) and the other for emergent intelligence (at the cost of data and computation). In this talk, I will argue why it is better for robots to be in the middle of this broad spectrum. Using manipulation as a case study, I will discuss how our lab combines ideas from dynamical systems and machine learning to overcome three often-overlooked issues with contemporary methods: i) high barrier to entry due to demands for expensive computational resources and annotated data, ii) inability to handle new tasks without relying on significant user expertise (e.g., for reward or controller design, hyperparameter tuning, data collection and curation), and iii) unreliable behaviors due to inscrutable and unpredictable learned policies. Addressing these issues will enable robot learning to escape the confines of well-resourced research labs and positively impact the larger society.
PRESENTER
Harish Ravichandar is an Assistant Professor in the School of Interactive Computing and a core faculty member of the Institute for Robotics and Intelligent Machines (IRIM) at Georgia Institute of Technology. He directs the Structured Techniques for Algorithmic Robotics (STAR) Lab, where his team leverages ideas from dynamical systems and control to design structured robot learning algorithms that improve reliability, efficiency, and self-sufficiency of robots operating in unstructured human environments. His research is motivated by and pertinent to diverse applications, ranging from dexterous manipulation to multi-agent coordination. His work has been recognized by IEEE MRS Best Paper Award, ASME DSCC Best Student Paper Award, IEEE CSS Video Contest Award, UTC Institute for Advanced System Engineering Graduate Fellowship, and Georgia Tech’s College of Computing Outstanding Post-Doctoral Research and Outstanding Research Scientist Awards.
zhlédnutí: 153
Video
Spring 2024 GRASP SFI - Karl Pertsch, University of California, Berkeley and Stanford University
zhlédnutí 222Před 21 dnem
“Building Open-Source Generalist Robot Policies” ABSTRACT Generalist robot policies, trained on large and diverse robot datasets, have the potential to transform how robot learning research is done: in the same way that current models in NLP are almost universally derived from pretrained large language models, future robot policies might be initialized from generalist robot models and finetuned...
Spring 2024 GRASP SFI - Michel Hidalgo, Ekumen
zhlédnutí 308Před měsícem
“Doing robotics in digital labs: Or how simulations fuel robotics development” ABSTRACT How do you do robotics without robots? Ekumen has been profitably providing software consulting services to robotics companies for over a decade, 10000 miles from relevant technological hubs. While multi-causal, a non-negligible factor in the company’s success were the advancements in multi-body dynamics sim...
Spring 2024 GRASP SFI Madhur Behl, University of Virginia, “Bringing AI Up To Speed”
zhlédnutí 267Před měsícem
ABSTRACT Why has autonomous driving, a task demanding significant intelligence, not met the high expectations set by many? Which hurdles have turned out to be more formidable than expected, and how can we refine our testing methodologies for autonomous vehicles (AVs) to address these problems more efficiently? In this talk, I will discuss the targeted research initiatives we have engaged in to ...
Spring 2024 GRASP SFI Eric Jang, 1X Technologies, “Data Engines for Humanoid Robots”
zhlédnutí 2,7KPřed měsícem
ABSTRACT 1X’s mission is to create an abundant supply of physical labor through androids that work alongside humans. I will share some of the progress 1X has been making towards general-purpose mobile manipulation. We have scaled up the number of tasks our androids can do by combining an end-to-end learning strategy with a no-code system to add new robotic capabilities. Our Android Operations t...
GRASP Lab Profile: Jasleen Kaur Dhanoa
zhlédnutí 222Před měsícem
Penn Engineering’s GRASP Lab fosters the development of inspiring leaders in cutting-edge robotics research. Jasleen Dhanoa completed the Robotics Master's Program within the School of Engineering & Applied Science at Penn. The Robotics MSE program provided her with a flexible curriculum which allowed her to tailor her course of study based on her interests. In May 2023, Jasleen graduated with ...
Spring 2024 GRASP SFI Eugene Vinitsky, New York University and Apple
zhlédnutí 95Před měsícem
“Real-world reinforcement learning in multi-agent systems: deploying cooperative autonomy at scale” ABSTRACT The ever-increasing penetration of level-2 autonomous vehicles (AVs) offers an opportunity to reshape the energy efficiency and throughput of our highways. Even at current low penetration rates (1-5%), we have observed in small settings that adopting different driving behaviors from huma...
Spring 2024 GRASP SFI Joseph DelPreto, Massachusetts Institute of Technology
zhlédnutí 116Před měsícem
There was a problem with the zoom recording, a small part of the end and the Q&A are missing. Sorry for the inconvenience. “Using Sensing and AI to Enrich Human Interactions with Machines and Nature” ABSTRACT Coupling advanced wearable and environmental sensors with dynamic AI frameworks has the potential to transform how we engage with machines and with the natural world. By co-developing inte...
Spring 2024 GRASP Seminar Saurabh Gupta, University of Illinois at Urbana Champaign
zhlédnutí 250Před 2 měsíci
Spring 2024 GRASP Seminar Saurabh Gupta, University of Illinois at Urbana Champaign
Spring 2024 GRASP Seminar Yutong Bai, Johns Hopkins University
zhlédnutí 693Před 2 měsíci
Spring 2024 GRASP Seminar Yutong Bai, Johns Hopkins University
Spring 2024 GRASP SFI Spring Berman, Arizona State University
zhlédnutí 109Před 2 měsíci
Spring 2024 GRASP SFI Spring Berman, Arizona State University
Spring 2024 GRASP SFI - Erik Bekkers, University of Amsterdam
zhlédnutí 118Před 2 měsíci
Spring 2024 GRASP SFI - Erik Bekkers, University of Amsterdam
Spring 2024 GRASP SFI - Fei Miao, University of Connecticut
zhlédnutí 119Před 2 měsíci
Spring 2024 GRASP SFI - Fei Miao, University of Connecticut
Spring 2024 GRASP SFI-Andrew Owens, University of Michigan, “Multimodal Learning from the Bottom Up”
zhlédnutí 106Před 2 měsíci
Spring 2024 GRASP SFI-Andrew Owens, University of Michigan, “Multimodal Learning from the Bottom Up”
GRASP Lab Profile: Abriana Stewart-Height
zhlédnutí 311Před 3 měsíci
GRASP Lab Profile: Abriana Stewart-Height
Fall 2023 GRASP SFI Ge Yang, NSF Institute of AI and Fundamental Interactions and MIT CSAIL
zhlédnutí 331Před 5 měsíci
Fall 2023 GRASP SFI Ge Yang, NSF Institute of AI and Fundamental Interactions and MIT CSAIL
Fall 2023 GRASP Seminar - Shangzhe Wu, Stanford University
zhlédnutí 490Před 5 měsíci
Fall 2023 GRASP Seminar - Shangzhe Wu, Stanford University
Fall 2023 GRASP on Robotics Nancy Amato, University of Illinois at Urbana Champaign
zhlédnutí 189Před 5 měsíci
Fall 2023 GRASP on Robotics Nancy Amato, University of Illinois at Urbana Champaign
Fall 2023 GRASP SFI - Xiaolong Wang, University of California San Diego
zhlédnutí 527Před 5 měsíci
Fall 2023 GRASP SFI - Xiaolong Wang, University of California San Diego
Fall 2023 GRASP SFI - Samuel Sokota, Carnegie Mellon University
zhlédnutí 196Před 5 měsíci
Fall 2023 GRASP SFI - Samuel Sokota, Carnegie Mellon University
Fall 2023 GRASP SFI - Margaret Coad, University of Notre Dame
zhlédnutí 152Před 5 měsíci
Fall 2023 GRASP SFI - Margaret Coad, University of Notre Dame
Fall 2023 GRASP Seminar - Daniel Gehrig, University of Zurich
zhlédnutí 201Před 5 měsíci
Fall 2023 GRASP Seminar - Daniel Gehrig, University of Zurich
Fall 2023 GRASP SFI E Farrell Helbling, Cornell University, “Autonomy for insect scale robots”
zhlédnutí 468Před 6 měsíci
Fall 2023 GRASP SFI E Farrell Helbling, Cornell University, “Autonomy for insect scale robots”
Fall 2023 GRASP - SFI Matthew D Kvalheim, University of Maryland, Baltimore County
zhlédnutí 128Před 6 měsíci
Fall 2023 GRASP - SFI Matthew D Kvalheim, University of Maryland, Baltimore County
Fall 2023 GRASP SFI Helmut Hauser, University of Bristol
zhlédnutí 136Před 6 měsíci
Fall 2023 GRASP SFI Helmut Hauser, University of Bristol
Fall 2023 GRASP SFI Andy Zeng, Google DeepMind, “From words to actions”
zhlédnutí 730Před 6 měsíci
Fall 2023 GRASP SFI Andy Zeng, Google DeepMind, “From words to actions”
Fall 2023 GRASP SFI Robert Baines, ETH Zürich
zhlédnutí 220Před 6 měsíci
Fall 2023 GRASP SFI Robert Baines, ETH Zürich
Fall 2023 GRASP SFI: David Lentink, University of Groningen, “Avian Inspired Design”
zhlédnutí 298Před 7 měsíci
Fall 2023 GRASP SFI: David Lentink, University of Groningen, “Avian Inspired Design”
Fall 2023 GRASP Seminar GRASP Research Overview - Day 2
zhlédnutí 543Před 7 měsíci
Fall 2023 GRASP Seminar GRASP Research Overview - Day 2
Fall 2023 GRASP Seminar: GRASP Research Overview - Day 1
zhlédnutí 611Před 7 měsíci
Fall 2023 GRASP Seminar: GRASP Research Overview - Day 1
They've developed a full body 3d screen now that doesn't require the polarized 3D glasses.
This is really cool
I'm the only Robot Dr. No we don't hang Robots specializing rusty joints .😊specializing in broken neck..Raymond Dr Ray
seem to be the better solution for difficult terrains cheaper and better autonomy of all legged solution and far more efficient than tracks with a lowest speed sole requierement finding the good bldc motor
It's amazing to see the progress in just over a decade!
What actuator is the tetrahedron robot uses
this research is amazing!
Edward Keonjian
can you drop the name of open source package here? so we can download. thanks.
Correction in 20:48. CMA-ME optimizes a neural network of 109 parameters and there are 1.42*10^(35) possible decks in Hearthstone.
A crowdfunding for your robotic project will work fine.
cant wait for next season.
Nice
Physical "intelligence" aka correct mechanical construction?
Not really! A physically intelligent system is wise about controlling its mechanical impedance (think how springy and energy-dampy my system needs to be) to match the scenario (mostly collision-related) it is faced with. Hence, there is no correct way! There is an old way that works well with application-specific machines in controlled environments like rigid robots in a factory setting. Some of the newer explorative, mapping and outdoor applications need robots to be more collision adaptive, hence the robotics paradigm shift.
That's a really nice talk! Thanks for sharing!
This is amazing! I'm starting to look into robotics and ai, and this is a great foundation! (Note to self, 45:59 is when the software part is mentioned)
Still interesting to watch after 11 years
♥.♥ Thanks. I think the idea for this type of wheel may have originally been related to that Soviet-designed tank called the "Soviet TPP-2 Jumping Tank".
Great review !! Thanks a lot !!
good video, want to be youtube friends?
We will watch your career with a great interest
very cool! the lucidity of this presentation is amazing
starts at 27:51 actually
You misspelled converging. Conversing on an object means they are speaking or otherwise carrying on a conversation on the object.
great work man , and I'm also interested what kind of motors are you using?
this lecture needs a transcript. It's a good lecture!
speak slowly. You are very hard to understand
I can't wait to see that in robot combat. Like Battle bots, Robot Wars and/or Robo Games. Heh I laughed when it slid down those stairs. Adored it when it was pronking and tracking that ball.
A little slow but awesome.
1000101110001011110000010000001101101111001101101000111011111000120000100010000111101011100111010101110
several years already and i am still in love with this robot.
Get back to you when i upload a video :P... probably be done b4 then end of the year
lolwut pics/dox or gtfo
Not to be an ass but I'm building far more complex robots at my house, with a current income of $0. I wouldn't be surprised if i could get a militarily contract of well over a few million if these guys can get million dollar militarily contracts. Haha remember my name so when I'm a world leader you can say I remember him when he was just getting out there. :)
Cute! He looks like a little dog or something :D Now imagine it with big sharp teeth and it's chasing you.. in an abandoned building.. at night.. when you're alone.
Terrible weapon or mans best friend? Only time will tell.
One of my favorite robots. :)
なんだか可愛いなぁ
This doesn't seems to be very energy efficient, especially if you consider a heavier object to carry... But nice work!
@FlandreScarletTepes First song is "In Space" by the group "Röyksopp", album name "Melody A.M.".
Not only are you geniuses, but you guys have excellent taste in music. I'd be delighted if you could give me any info about either of the songs.
Biped mode is just brilliant. :)
i think i've fallen in love c:
it scored by chance... XD
why aren't they playing Tetris yet ?!?! :oP
Player/Stage/Gazebo
what program did you use?
Cute!
Pretty cool! Now, this wouldn't improve the science & engineering behind this project, but have you considered tiny Dalek-shaped shells for these things? ;)
Excellent! :)