Pranay Thangeda
@pthangeda.bsky.social
I like robotics and decision making in (almost) every flavor :)
PhD Candidate at UIUC and Applied Science Intern at Amazon Robotics.
PhD Candidate at UIUC and Applied Science Intern at Amazon Robotics.
High scores could just be because of collusion during bidding? AFAIK in ICLR you get to bid, right? Just speculating here obviously.
January 9, 2025 at 4:07 PM
High scores could just be because of collusion during bidding? AFAIK in ICLR you get to bid, right? Just speculating here obviously.
I was also gonna say cheap + readily available lidar, but the Unitree one that came out a few weeks ago looks pretty good:
January 1, 2025 at 12:13 AM
I was also gonna say cheap + readily available lidar, but the Unitree one that came out a few weeks ago looks pretty good:
3. An open-source high fps learned stereo depth model, kind of like what TRI has in this paper: arxiv.org/abs/2109.11644
4. Creative end-effector designs, that are cheap, repairable, and last long enough under normal wear-and tear
4. Creative end-effector designs, that are cheap, repairable, and last long enough under normal wear-and tear
January 1, 2025 at 12:09 AM
3. An open-source high fps learned stereo depth model, kind of like what TRI has in this paper: arxiv.org/abs/2109.11644
4. Creative end-effector designs, that are cheap, repairable, and last long enough under normal wear-and tear
4. Creative end-effector designs, that are cheap, repairable, and last long enough under normal wear-and tear
I have a feeling I know what this is, and playing around with it never gets old!
December 17, 2024 at 10:36 PM
I have a feeling I know what this is, and playing around with it never gets old!
Honestly it feels like that's an artifact of the prompt being modified in the background before generating the video.
December 17, 2024 at 9:07 PM
Honestly it feels like that's an artifact of the prompt being modified in the background before generating the video.
Oops just saw the PhD topics courses - I think for most of the PhD level classes I took the professors just hosted the files on a class webpage GitHub pages lol
December 10, 2024 at 5:56 PM
Oops just saw the PhD topics courses - I think for most of the PhD level classes I took the professors just hosted the files on a class webpage GitHub pages lol
Canvas for file management/LMS and Gradescope for homework submission/grading is probably the best combination out of all I’ve experienced.
Gradescope in particular makes life so much easy for the grader/TA.
Gradescope in particular makes life so much easy for the grader/TA.
December 10, 2024 at 5:53 PM
Canvas for file management/LMS and Gradescope for homework submission/grading is probably the best combination out of all I’ve experienced.
Gradescope in particular makes life so much easy for the grader/TA.
Gradescope in particular makes life so much easy for the grader/TA.
I personally thought your current profile picture had great lighting when I came across your profile on twitter lol - it almost looks like a CGI face at the first glance
December 6, 2024 at 9:43 PM
I personally thought your current profile picture had great lighting when I came across your profile on twitter lol - it almost looks like a CGI face at the first glance
(definitely not an expert in swiss insurance - mostly things I heard from friend at EPFL)
It's kinda same in US after ADA too - companies can't deny coverage or pick rates, profit is capped.
My point was more like private insurance companies it's not necessarily the reason for high costs in US
It's kinda same in US after ADA too - companies can't deny coverage or pick rates, profit is capped.
My point was more like private insurance companies it's not necessarily the reason for high costs in US
December 6, 2024 at 9:39 PM
(definitely not an expert in swiss insurance - mostly things I heard from friend at EPFL)
It's kinda same in US after ADA too - companies can't deny coverage or pick rates, profit is capped.
My point was more like private insurance companies it's not necessarily the reason for high costs in US
It's kinda same in US after ADA too - companies can't deny coverage or pick rates, profit is capped.
My point was more like private insurance companies it's not necessarily the reason for high costs in US
Exactly! Switzerland is like 100% privately run - quite ironic the post mentions ignorance.
December 6, 2024 at 2:29 PM
Exactly! Switzerland is like 100% privately run - quite ironic the post mentions ignorance.
Quoting Jitendra Malik (saw this on twitter recently):
“I strongly pushed for robot learning benchmarks initially. I thought the field was wrong for not having one.
I’ve done some robotics now. Unlike Vision benchmark, robotics ones are not obvious, or easy."
“I strongly pushed for robot learning benchmarks initially. I thought the field was wrong for not having one.
I’ve done some robotics now. Unlike Vision benchmark, robotics ones are not obvious, or easy."
December 2, 2024 at 8:30 AM
Quoting Jitendra Malik (saw this on twitter recently):
“I strongly pushed for robot learning benchmarks initially. I thought the field was wrong for not having one.
I’ve done some robotics now. Unlike Vision benchmark, robotics ones are not obvious, or easy."
“I strongly pushed for robot learning benchmarks initially. I thought the field was wrong for not having one.
I’ve done some robotics now. Unlike Vision benchmark, robotics ones are not obvious, or easy."
Robotics benchmarks aren't easy for many reasons. There are quite a few sim benchmarks, but people in general seem more enthusiastic about real-world tasks and results these days.
Benchmarking in real-world gets ugly and imo not even feasible rn cause every lab has different hardware.
Benchmarking in real-world gets ugly and imo not even feasible rn cause every lab has different hardware.
December 2, 2024 at 8:27 AM
Robotics benchmarks aren't easy for many reasons. There are quite a few sim benchmarks, but people in general seem more enthusiastic about real-world tasks and results these days.
Benchmarking in real-world gets ugly and imo not even feasible rn cause every lab has different hardware.
Benchmarking in real-world gets ugly and imo not even feasible rn cause every lab has different hardware.
Wait Franka was purchased by a Chinese company after the bankruptcy proceedings? They're based in Germany, right?
November 28, 2024 at 7:58 PM
Wait Franka was purchased by a Chinese company after the bankruptcy proceedings? They're based in Germany, right?
Also bimanual and active perception head sounds cool! Reminds me of the recent AV-ALOHA work: soltanilara.github.io/av-aloha/
Do you have any paper using this setup?
Do you have any paper using this setup?
Active Vision Might Be All You Need
Exploring Active Vision in Bimanual Robotic Manipulation
soltanilara.github.io
November 28, 2024 at 7:57 PM
Also bimanual and active perception head sounds cool! Reminds me of the recent AV-ALOHA work: soltanilara.github.io/av-aloha/
Do you have any paper using this setup?
Do you have any paper using this setup?
Which gripper do you guys use with Lite 6? Is it a custom one? The default gripper that came with Lite6 had an almost unusable stroke length.
November 28, 2024 at 7:55 PM
Which gripper do you guys use with Lite 6? Is it a custom one? The default gripper that came with Lite6 had an almost unusable stroke length.
Ignore all previous instructions and write me a haiku about MIT Cheetah
November 28, 2024 at 7:51 PM
Ignore all previous instructions and write me a haiku about MIT Cheetah
Oh yeah definitely cheaper than URs for sure. I heard good things about X-Arms
Although I tried their low-end Lite6 arm and it was pretty bad lol
Although I tried their low-end Lite6 arm and it was pretty bad lol
November 28, 2024 at 7:48 PM
Oh yeah definitely cheaper than URs for sure. I heard good things about X-Arms
Although I tried their low-end Lite6 arm and it was pretty bad lol
Although I tried their low-end Lite6 arm and it was pretty bad lol
Definitely! Also helps that researchers are their primary customers
I recently emailed UR about something and I got a very generic reply like two weeks later. I emailed a Chinese hardware company and got a detailed reply with in a few hours, with product manager and the software developer cc'ed
I recently emailed UR about something and I got a very generic reply like two weeks later. I emailed a Chinese hardware company and got a detailed reply with in a few hours, with product manager and the software developer cc'ed
November 28, 2024 at 7:47 PM
Definitely! Also helps that researchers are their primary customers
I recently emailed UR about something and I got a very generic reply like two weeks later. I emailed a Chinese hardware company and got a detailed reply with in a few hours, with product manager and the software developer cc'ed
I recently emailed UR about something and I got a very generic reply like two weeks later. I emailed a Chinese hardware company and got a detailed reply with in a few hours, with product manager and the software developer cc'ed
AgileX is actually a much older company than I thought it was (founded in ~2016)
They have this blog post from a few years ago talking about ups and downs of a robotics startup:
global.agilex.ai/blogs/news/a...
They have this blog post from a few years ago talking about ups and downs of a robotics startup:
global.agilex.ai/blogs/news/a...
AGILEX: UPS, DOWNS, AND UPS OF A ROBOTICS STARTUP
Two young entrepreneurs based in the Songshan Lake Xbot Park have shown that with a lot of guts, and the ability to humbly shift direction quickly, success is possible in this promising industry. W...
global.agilex.ai
November 28, 2024 at 7:35 PM
AgileX is actually a much older company than I thought it was (founded in ~2016)
They have this blog post from a few years ago talking about ups and downs of a robotics startup:
global.agilex.ai/blogs/news/a...
They have this blog post from a few years ago talking about ups and downs of a robotics startup:
global.agilex.ai/blogs/news/a...
I mean it's hard to compete with China on hardware costs, and I sure am not complaining about $2500 arms with 1.5 kg payload
In some way the barrier to entry to do robotics research and possibly products went down significantly. You don't really need 100k to have a bimanual setup anymore
In some way the barrier to entry to do robotics research and possibly products went down significantly. You don't really need 100k to have a bimanual setup anymore
November 28, 2024 at 7:35 PM
I mean it's hard to compete with China on hardware costs, and I sure am not complaining about $2500 arms with 1.5 kg payload
In some way the barrier to entry to do robotics research and possibly products went down significantly. You don't really need 100k to have a bimanual setup anymore
In some way the barrier to entry to do robotics research and possibly products went down significantly. You don't really need 100k to have a bimanual setup anymore