Pinned Tweet
Introducing STERLING 💫: "Self-Supervised Terrain Representation Learning from Unconstrained Robot Experience" 💫 to be presented at CoRL 2023 in Atlanta! 😀
Project Page 🌎: hareshkarnan.github.io/sterl…
Paper 🗞️: openreview.net/pdf?id=VLihM6…
A thread 🧵
Haresh Karnan retweeted
I’m hiring multiple PhD students and one postdoc at NTU Singapore, starting in Spring or Fall 2026, to push the frontiers of robot planning, learning, and embodied AI.
Details are available here: yoonchangsung.com/opportunit…
Thank you for your support in sharing this opportunity!
#PhDOpportunity #Robotics #NTUsg
Haresh Karnan retweeted
What's green, bidirectional, and now driving autonomously in San Francisco and the Las Vegas Strip? The Zoox robotaxi! Give us a wave if you see us on the road! 👋
Head to our Journal for all the details: zoox.com/journal/zoox-robota….
#Robotaxi
Haresh Karnan retweeted
Excited to announce the call for workshop proposals for @RL_Conference is open!
rl-conference.cc/call_for_wo…
Due March 8th.
@JosiahHanna and I are the workshop co-chairs this year and we hope you'll consider submitting a proposal. Workshops on any RL-related topics are welcome!
Targeted Learning will be presented at #ICRA24! 🎉 Checkout the paper and video if you're interested in social navigation!
Replying to @XuesuXiao
3. Targeted Learning: A Hybrid Approach to Social Robot Navigation, by Amir Hossain Raj, Zichao Hu, @KarnanHaresh, @rohanchandra30, Amir Payandeh, Luisa Mao, @PeterStone_TX, @Joydeepb_robots, and @XuesuXiao. Paper: cs.gmu.edu/~xiao/papers/hybr… Video: youtube.com/watch?v=QT0JTuPh…
Haresh Karnan retweeted
Apparently, Ingenuity flew over a texture-poor terrain and RANSAC wasn’t able to find inliers: spectrum.ieee.org/mars-helic… According to NASA: “It's very featureless-bland, sandy terrain. And that's why we believe that during Flight 71, we had an emergency landing. She was flying over the surface and was realizing that there weren't too many rocks to look at or features to navigate from, and that's why Ingenuity called an emergency landing on her own." “The way a system like this works is by looking at the consensus of [the features| it sees, and then throwing out the things that don't really agree with the consensus. The danger is when you run out of features, when you don't have very many features to navigate on, and you're not really able to establish what that consensus is and you end up tracking the wrong kinds of features, and that's when things can get off track.”
Here's everything @NASAJPL knows about what likely happened on Ingenuity's last flight spectrum.ieee.org/mars-helic… #MarsHelicopter
Pretty cool. Mixtral-8x7b gets the same Arena Elo rating as GPT-3.5T, a first of its kind with a non-proprietary license
[Arena Update]
We've collected over 6000 and 1500 votes for Mixtral-8x7B and Gemini Pro. Both show strong performance against GPT-3.5-Turbo.
Big congrats again on the release! @MistralAI @GoogleDeepMind
Full leaderboard: huggingface.co/spaces/lmsys/…
2
350
The new Mixtral-8x7b model is impressive! From my initial observations, coding capabilities seem just as good as GPT3.5, however, it seems to struggle with longer chat sequences where it tends to repeat "I'm not sure what you mean by.. "
141
Haresh Karnan retweeted
🌟Excited to present our paper (arxiv.org/abs/2306.09526) at #NeurIPS2023 next week! We introduce a novel RL framework, Residual Q-learning, to customize an imitative policy towards additional task requirements. Join us for the poster session on 13 Dec, 10:45 a.m. CST! 🧵
Impressive!
Introducing Pika 1.0, the idea-to-video platform that brings your creativity to life.
Create and edit your videos with AI.
Rolling out to new users on web and discord, starting today. Sign up at pika.art
184
Haresh Karnan retweeted
Thrilled to announce the first annual Reinforcement Learning Conference @RL_Conference, which will be held at UMass Amherst August 9-12!
RLC is the first strongly peer-reviewed RL venue with proceedings, and our call for papers is now available: rl-conference.cc.
🚨This is tomorrow! Remember to stop by the poster session from 12-12:45 to learn more :)
Introducing STERLING 💫: "Self-Supervised Terrain Representation Learning from Unconstrained Robot Experience" 💫 to be presented at CoRL 2023 in Atlanta! 😀
Project Page 🌎: hareshkarnan.github.io/sterl…
Paper 🗞️: openreview.net/pdf?id=VLihM6…
A thread 🧵
Looking forward to presenting STERLING @ #CoRL23 in Atlanta next week ! More details in the thread below, do visit our poster session on Wednesday to learn more ! 😀
Also, if you happen to attend CoRL, feel free to dm. Happy to chat and connect with fellow roboticists! :)
Introducing STERLING 💫: "Self-Supervised Terrain Representation Learning from Unconstrained Robot Experience" 💫 to be presented at CoRL 2023 in Atlanta! 😀
Project Page 🌎: hareshkarnan.github.io/sterl…
Paper 🗞️: openreview.net/pdf?id=VLihM6…
A thread 🧵
Haresh Karnan retweeted
More good coverage of our robot parade from a couple of weeks ago:
news.utexas.edu/2023/10/16/s…
@texas_robotics
👻👻
My students took 2nd place in the TRGS (Texas Robotics Graduate Students) robot costume competition.
They modified the software to be "Spooky" for Halloween! Check out the video of our haunted robot!
youtu.be/zrX4Pq5A8fQ
#livingwithrobotslab
@PeterStone_TX @texas_robotics
1
151
Interesting work from @agsidd10 on f-Policy gradients to learn optimal goal-conditioned policies with goal distribution matching. Check it out !
In domains with sparse rewards, reward shaping is well known to speed up learning by providing a dense learning signal.
We introduce an alternate method, f-Policy Gradients (arxiv.org/abs/2310.06794v1), to obtain optimal policies through distribution matching. (1/n)
#NeurIPS2023
1
160
Interested in urban context-aware navigation ? Checkout 💫 CODa 💫, the latest and largest multi-modal, annotated robot dataset from Texas Robotics led by @ArthurKZhang
Context-aware navigation requires object/terrain-level understanding - in support of this, we are thrilled to release CODa, the largest multi-class, multi-modal urban robotics dataset to date.
amrl.cs.utexas.edu/coda/
Work done with a fantastic team from @amrl_ut!
A short 🧵, 1/7
4
299
Haresh Karnan retweeted
Tasks at home can involve many objects with complex interdependencies, e.g., the butter needs to be melted on the stove before it can be mixed and fried with meat in a pot.
How can RL efficiently explore such tasks with sparse rewards?
Introducing our #NeurIPS23 paper ELDEN: 🧵
Haresh Karnan retweeted
Dobby is a conversational service robot and the starting point of my lab's new work on LLMs in robotics!
arxiv.org/abs/2310.06303
youtu.be/iAuRsHsZMeM
#livingwithrobotslab @PeterStone_TX @texas_robotics @UTCompSci

























