It’s not hard to see
@ElonMusk
’s logic;
@Tesla
as a robotics company could justify much higher P/E than Tesla as a car company. But robotics isn’t rocket science — it’s harder. Glad to have
@Tesla
join the effort!
I get asked a lot: why stay in academia, all the excitement in AI is happening in industry with massive compute. And I am seeing some profs leaving academia, but also seeing lots of researchers in industry looking to go back to academia, especially those who don’t work on LLMs.…
The earth has experienced 5 mass extinctions where life was sustained under the oceans. Isn’t that a better bet than the moon or Mars if we one day need to take refuge?
Deep Learning Can Significantly Accelerate Grasp-Optimized Motion Planning. Results can reduce computation time for a pick-and-place task from 29 seconds to 80 milliseconds. Science Robotics
@AAAS
Excited to announce that
@Jeff_Ichnowski
, a stellar robotics and systems postdoc in the
@AUTOLab_Cal
, originator of Grasp Optimized Motion Planning (GOMP) and Cloud Robotics for ROS (
#FogROS2
), will start as Asst. Prof. of Computer Science
@CarnegieMellon
in January!
197 papers accepted for Conference on Robot Learning in Auckland (39% acceptance rate, same as last year). Major thanks to Program Chairs and Reviewers.
@corl_conf
#CoRL22
.
A high-quality dataset of 696 common household objects that
@GoogleResearch
3D-scanned is now available under
@creativecommons
license for researchers in robot simulation and perception:
The
@IEEE
Robotics and Automation Society just voted to allocate USD$50,000/year for their conferences to provide services requested by attendees with disabilities (eg sign-language interpreters). Kudos to Torsten Kroeger, Daniel Seita, and others for leading this effort.
I ❤️ editing. Especially drafts by the brilliant, dedicated students in my lab.
#AUTOLab
@UCBerkeley
(good luck to them + all racing to make the
#ICRA2020
deadline tonight!)
Today at 5:10pm PST: Talk by
#ChatGPT
lead engineer and
@UCBerkeley
alum John Schulman: Reinforcement Learning from Human Feedback: Progress and Challenges. Berkeley AI Research Lab
@berkeley_ai
@Berkeley_EECS
Livestreamed:
The Conference on Robot Learning 2022 website is now up!
@corl_conf
Submission deadline 15 June. Happy New Year (although it's already 2022 in New Zealand;):
The word "
#robot
" was coined in 1920. A small word with a gigantic influence on human imagination (I may be slightly biased). A toast to the robots of the past century and those ahead...
Very interesting to hear
@geoffreyhinton
questioning backpropagation, advocating for local reward functions, describing adversarial examples as human error, and recounting the history of t-SNE! (BTW: skeptics are quite different than deniers;)
The first two are based on recent moves at
@Meta
and
@DeepMind
. Robotics&Automation are attracting interest from many sectors and smart VCs. The key is not to expect miracles but to hunker down: demonstrate new prototypes and engineer them into reliable, cost-effective systems.
As
@Animesh_Garg
noted, the future of robot-assisted surgery is not about replacing surgeons. Like cruise control for cars, it’s about increasing consistency and decreasing tedium.
@CadeMetz
’s article describes our new results.
All videos of keynotes and talks from 1st Conference on
#Robot
Learning (
#CoRL2017
) incl. keynotes by
@rodneyabrooks
,
@ylecun
,
Stefanie Tellex, Anca Dragan, and Drew Bagnell and 50 papers and Open Problems Session notes:
Our warmest congratulations to Pieter Abbeel
@pabbeel
, recipient of the 2021
#ACMPrize
for contributions to robot learning, including learning from demonstrations and deep reinforcement learning for robotic control.
Learn more about Abbeel’s work here:
ML legend Michael Jordan is moving away from the singular intelligence paradigm of AI and focusing on complementarity—AI that enhances the productivity of human groups:
Looking fwd to presenting a new talk: "Data is All You Need: Large Robot Action Models and Good Old Fashioned Engineering"
@Stanford
Robotics Seminar on Friday:
CoRL: 6th Conference on Robot Learning will be in Auckland, New Zealand, 14-18 Dec 2022 (where it'll be summertime ;):
@CoRL_Conf
Deadline 15 June: We welcome submissions addressing the theory and practice of machine learning for robots and automation.
To hire top engineers
@elonmusk
needs ambitious visions. Optimus is similar to robots developed by
@BostonDynamics
and
@agilityrobotics
. It's not clear how it will perform reliable manipulation and be cost effective. But hats off to the Great Muskini for getting our attention!
I caught it a few weeks ago from a friend and then passed in on to several family members. No one seems able to shake it. I think everyone is going to end up getting infected. We’re just going to have to learn how to live with
#Wordle
.
The 2020 Bay Area Robotics Symposium (BARS) on Friday included a fascinating variety of new results summarized in short talks ( ) all available now online, including keynote from
@RodneyaBrooks
:
Remembering building a robot with my father, Melvin Morris Goldberg. An engineer, inventor, and entrepreneur with an irreverent sense of humor, he passed away on Dec 9, 1982.
Tangled cables are unwieldy, unsightly, and a tripping hazard in homes and workplaces – including retail, factories, boats, and rock concerts ;). Can robots help? Thread 🧵👇 (1/7)
505 papers received! Thanks to all authors. Program Committee headed by Karen Liu,
@ProfKulic
,
@Jan_R_Peters
, and
@jeff_ichnowski
are working now with Area Chairs to assign reviewers!
Today is the day! Looking forward to your submissions!
#CoRL2022
🤖
Have you already tested our submission system and added an early version? Time to try it out! 😉
#robotics
#machinelearning
#1
topic requested by
@ycombinator
is
#Robotics
: "Robotics hasn't yet had its GPT moment, but we think it’s close.... We're interested in people building software tools to help other people to make robots:
Why do artists often over-estimate the skills of engineers, while engineers often under-estimate the skills of artists? It would be ideal if we could eliminate both biases.
"The amount of hard work and studying that’s needed to learn how to be an artist is analogous to the hard work that we’re well aware of in engineering" -
@UCBerkeley
prof.
@ken_goldberg
, in
#NAEBridge
article on
#engineering
and art.
Read more here:
Deeply saddened by this news. JPL was a brilliant friend and colleague; a pioneer in robot motion planning and thoughtful thinker about art, culture, and robots. RIP.
The
@Yale
“Model W” robot gripper uses palm and link contacts for dextrous manipulation. It was designed by optimizing joint parameters using a metric inspired by caging theory.
A
#robotic
hand equipped with a palm was able to perform a variety of contact-rich hand motions - for example, fidgeting with objects - that are difficult for most robotic hands equipped only with fingers. Designed by
@YaleSEAS
scientists:
Seeking Postdoc with experience in Robotics and Cloud-Based Systems (AWS, Google Cloud, Azure) to work closely with
@UCBerkeley
Profs. Ken Goldberg and Joey Gonzales on research on cloud robotics architectures and platforms such as FogROS2:
The first important fact to realize when thinking about AI is that an agent in physical reality literally NEVER encounters the same input state twice. The second is to realize that two very similar input states may require vastly different actions based on some small detail.
@UofTRobotics
is hosting our first live-streamed seminar!
Catch
@ken_goldberg
discussing the "New Wave in Robot Grasping" on Friday, March 19th at 1pm Eastern on our YouTube page:
A robot can uniquely orient any polygonal part without sensing, using an algorithm from 1991. A new interactive website lets you test it with your own part shapes:
Very proud of the team
@AmbiRobotics
for extending Dex-Net grasping research from
@UCBerkeley
to achieve Superhuman Sorting -- picking, scanning, and placing -- a wide range of real packages at twice the speed of humans.
Introducing AmbiSort, an
#AI
-powered
#robotic
sorting line, powered by AmbiOS, our advanced simulation-to-reality AI operating system that allows associates to work at the speed of
#ecommerce
. Learn more at
Relax everyone! This headline is greatly overstated. This is a mild improvement of work reported in 2016 where a robot with specialized lighting and a suture gun positions the gun. The results are interesting but researchers are very far from putting surgeons out of work.
You can relax. Claims about AGI are greatly exaggerated. There are real dangers for our planet: climate change, nuclear war, virus mutations, but AI is not one of them.
Call For Papers: CoRL: 6th Conference on Robot Learning will be in Auckland, New Zealand, 14-18 Dec 2022 (online and in-person during summertime ;):
@CoRL_Conf
Deadline 15 June:
Correct: "Robots are still quite clumsy compared with humans when dealing with physical manipulation in unstructured environments and tasks.": What can machine learning do? [
@erikbryn
@tommmitchell
I was a PhD student in the Leg Lab. That was 40 years ago. Marc and
@BostonDynamics
have made remarkable progress but this compilation left out the bloopers and it misses
@ylecun
’s point that progress in LLMs does not necc. imply progress in robotics.
In ML researchers often write about the "mismatch" between distributions, models, etc, but it seems the correct word is "disparity". Mismatch describes a discrete category error, whereas disparity is the degree of difference eg, KL divergence measures disparity, not mismatch.
But wait, this is not everything!
We now have ALL OUR TALKS ONLINE!
This includes keynotes, tutorials, orals, spotlights, sponsors, opening & closing sessions...
Trying to find keywords, authors, topics for the 1000+ papers at
#ICRA2020
? This spreadsheet includes links to the Slack channel for every paper listing the 10 min video, pdf, and Q&A thread:
Really like this "portrait" of our lab featuring things we work with including robot grippers, suction cups, vacuum valve, camera, adversarial grasp objects, assorted tools & coffee pod ;) by Adam Lau
@UCBerkeley
Larger version:
Computational neural networks are *very* loose approximations of biological brains. This is an excellent summary of key differences: by Matthew Roos in
@TDataScience
Many scientists view artists as soft and fuzzy, and artists view scientists as clueless about history and culture... But in both art and science, if you don’t deeply understand what has gone before, you can’t innovate:
I really appreciate the chance to discuss the design of robot grippers, the role of fault tolerance in automation applications, and the importance of dirty laundry for research in
@MarwaEldiwiny
’s exciting
@ieeeras
podcast series!
It was a pleasure to have Ken Goldberg
@ken_goldberg
on the podcast, he is one of my role models professors since I was undergrad student, I enjoyed this conversation on robotics grasping, I hope you find it useful.
#robotics
The episode
Just saw this and couldn’t peel my eyes away from it: stunning details and textures referencing art history. The most compelling generative work I’ve seen. Hats off to you Refik.
I’m deeply honored and excited to announce our exhibition — Unsupervised opening on the 19th of November at MoMA! It’s a dream of many artists to have the opportunity to imagine, practice and exhibit for such an important institution that supports the creativity of humanity!
With a Tiger in our tank, we’re hiring: “
@AmbiRobotics
has quickly leapt to the front of the pack when it comes to AI robotic piece-picking for parcel sortation and ecommerce fulfillment operations," said Griffin Schroeder, Partner, Tiger Global.
We are proud to announce the completion of our Series A
#funding
! The money raised enables us to boost deployments of our
#AI
-powered sorting
#robots
to our supply chain customers.
Read our press release here:
Very honored to see recognized in this way. The team of PhD students out of
@UCBerkeley
@Cal_Engineer
is commercializing and extending advances combining deep learning with sim2real for robot grasping and picking:
The Overall Excellence Awards go to ecommerce and logistics robotics companies
@AiAmbidextrous
and
@FetchRobotics
It was a very tough decision with so many other great companies to choose from! See all 52 stories at
Join us next Tuesday 12-1 PM (in-person + livestreamed) for a special panel on The Past and Future of Robotics and Machine Intelligence! Hear from 250+ years of combined experience w/ Ruzena Bacjsy,
@rodneyabrooks
,
@Ken_Goldberg
,
@JitendraMalikCV
, Shankar Sastry, Claire Tomlin.
Excited also about this BAIR blog comparing approaches to deep learning for robot manipulation of fabric, a hard problem with applications in hospitals, homes, retirement centers, and assembly lines.
Kudos to all who contributed to making this conference such a success! And a special thanks to my students who traveled from
@UCBerkeley
to present their papers:
My personal highlight of
#ICRA2023
!
ICRA was grand in every aspect, be it the exhibitions showcasing the state of the art robots, the amazing posters, and mind-blowing keynotes and plenary talks.
And interestingly...this is still just the beginning for robotics!
Thanks Rod. We don’t want to dampen the recent enthusiasm for Gen AI in robotics and humanoids, but it may be helpful for everyone to “curb our enthusiasm” a bit and remember GOFE so we can flatten the hype curve and avoid a bubble…
End to end learning for robotics is in a bit of a hype cycle but proponents sometimes miss the fact that practical results are not just end to end learning. As UC Berkeley Professor
@Ken_Goldberg
points out in this clip from a recent talk he gave at Stanford, Good Old Fashioned…
Then we are pretty far from “AGI”: Many jobs require dexterity. Advances in AI will contribute to robotics but the ability to solve math problems and even to discover new proofs and models does not imply the ability to design and construct efficient, reliable robots.
I like this definition. I’d raise the bar to 95% performance of 95% economically valuable jobs. This would necessarily require us to solve ROBOTICS. LLMs alone are not enough to realize physically embodied, generalist agents.
Robotics will be the last and by far the most…
@Jan_R_Peters
@animesh_garg
@ken_goldberg
because there are too many people claiming to do robot learning, while what they do is learn from simulated robots in a very simplistic simulated world. and they think if it works in simulation, then it will on the robot.
Starts in 15 mins!
@NationalSawdust
in Brooklyn NY presents: Breathless: Catie and the Robot, an eight-hour modern dance homage to physical labor.
Created by
@CatieCuan
and
@KenGoldberg
Saturday, December 16, 2023 (3 shifts: 2-10pm, discussion: 10-11pm):
A novel approach to integrating model-based and model-free RL using a few suboptimal demonstrations, by
@berkeley_ai
PhD students Ashwin Balakrishna and Brijen Thananjeyan with
@svlevine
and Joey Gonzalez and Francesco Borelli and others.