Robotics faculty @JHUCompSci | https://xitter.cf/t.co/K9F82bmmmk
Berkeley, CA
Joined March 2011
- Tweets1.3K
- Following2.1K
- Followers5.6K
- Likes6.2K
Pinned Tweet
This Fall, I'm teaching a new "Hands-on Robot Learning" class at @JHUCompSci
A full-stack class where students will get SO-101 robot kits, build robots out, collect data, train and deploy learned policies/WAMs/agents/etc.
Materials will be posted here:
hands-on-robot-learning.gith…
Krishna Murthy retweeted
🚨 Attention, Robot Learning community!
For the first time, @corl_conf is introducing a Call for Late-Breaking Results, focusing on Agentic Robotics and Real2Sim2Real!
Research is moving fast, and we want to make room for important new results as they emerge!
📅 Deadline: October 26 #CoRL2026
🔗 corl.org/contributions/call-…
Krishna Murthy retweeted
Out of curiosity, I left Claude overnight with the task of rewriting Ruckig (including the closed-source proprietary add-ons like intermediate waypoints) in modern C++.
It worked, it's slightly faster, and it let me make a 🇨🇱 reference. Wins all around.
github.com/sea-bass/charqui
Krishna Murthy retweeted
Excited to share 𝗕𝗥𝗔𝗖𝗘: a whole-body tracker for humanoids that can exert forces from diverse poses and adapt to varied terrains.
Humanoid whole-body trackers are typically trained to follow references recorded on flat ground, carrying nothing. As a result, they know 𝗻𝗲𝗶𝘁𝗵𝗲𝗿 𝘁𝗵𝗲 𝗳𝗼𝗿𝗰𝗲𝘀 𝘁𝗵𝗲 𝗿𝗼𝗯𝗼𝘁 𝗺𝘂𝘀𝘁 𝗲𝘅𝗰𝗵𝗮𝗻𝗴𝗲 𝘄𝗶𝘁𝗵 𝗼𝗯𝗷𝗲𝗰𝘁𝘀 𝗻𝗼𝗿 𝗰𝗮𝗻 𝘁𝗵𝗲𝘆 𝗮𝗱𝗮𝗽𝘁 𝘁𝗼 𝗱𝗶𝘃𝗲𝗿𝘀𝗲 𝘁𝗲𝗿𝗿𝗮𝗶𝗻𝘀. BRACE closes both gaps in a single tracker.
Our key innovation is a 𝗿𝗲𝗳𝗲𝗿𝗲𝗻𝗰𝗲-𝗮𝗱𝗮𝗽𝘁𝗮𝘁𝗶𝗼𝗻 𝗳𝗼𝗿𝗺𝘂𝗹𝗮𝘁𝗶𝗼𝗻. Given humanoid reference trajectories, we adapt them to the terrain and the commanded force, then train the controller on these adapted references. Tracking them naturally produces the stance and bracing needed to handle the terrain and exert the force (or compensate for external loads). 𝗧𝗵𝗲 𝗿𝗲𝘀𝘂𝗹𝘁𝗶𝗻𝗴 𝗰𝗼𝗻𝘁𝗿𝗼𝗹𝗹𝗲𝗿 𝗱𝗲𝗽𝗹𝗼𝘆𝘀 𝗱𝗶𝗿𝗲𝗰𝘁𝗹𝘆 𝗼𝗻 𝘁𝗵𝗲 𝗿𝗲𝗮𝗹 𝗿𝗼𝗯𝗼𝘁.
To know more, please visit our:
Project page: multiplylabs.github.io/brace…
Interactive demo: multiplylabs.github.io/robog…
More details in this thread🧵
1/n
Krishna Murthy retweeted
introducing claude-as-cad
i've been actually building robots and parts without knowing any CAD, using a claude-as-cad recipe -- here's how and you can too:
language-based prompting for CAD-agents is a good start, but it's imprecise -- these builds are difficult to steer exactly, among other issues
instead, you should have claude build you a claude-native CAD interface.
i've had success with this and found 3 important ingredients:
1. click to reference -- click any part of the build and the [3D coordinate, mesh face, object id] are copied to your clipboard to paste into chat
➡️ lets you be hyper-specific about what part you are referring to in chat
2. 3D primitive editor -- reposition/reorient any existing object; add a cube to the scene and scale/position for claude to use for reference
➡️ often times its hard to describe a position/transform in language and easier to illustrate it yourself to claude
3. hierarchical notes -- claude organizes separate builds into nested folders and writes skills that survive conversation compacting
➡️ saving different builds in folders with common skills, e.g. servo mounts, means once you get it right and save it, you can always reference it easily
link to install to your agent setup below ⬇️
Krishna Murthy retweeted
Introducing RoboJEPA, the first scaling law for multiembodiment world models.
We scaled JEPA predictors up to 8B params, showed scaling laws extrapolate and correlated compute scaling with planning performance.
We release all checkpoints and the code.
robojepa.github.io
Krishna Murthy retweeted
NEVER LOOK BACK!
Introducing LEDGER: persistent 3D object memory from egocentric videos.
Remember where things were, when they moved, and what they held -> then answer questions without looking back at the video!
ledger-3d.github.io
Krishna Murthy retweeted
Objects in robot demos are awfully patient 🙃
We’ve spent months on tasks where they aren’t. Catching balls, pouring into a moving pot, taking a bottle from someone walking by.
Two additions helped π0.5, RTC, & VLASH handle dynamic manipulation! More from our CoRL'26 oral in 🧵
Daniel & Jimin ran 1,500+ physical robot evaluations to put it to the test, with 50 runs per reported success rate.
As always, we’re making everything public: tempo-robot.github.io/
Krishna Murthy retweeted
Come work with us @JHUCompSci! We're hiring tenure-track faculty in all areas, with a particular focus on
- AI systems
- AI safety and security
- AI reasoning and alignment
- AI ethics, policy, and governance
apply.interfolio.com/194231
(DMs are open if you have any questions)
Krishna Murthy retweeted
1/6 "3D Vision is dead, long live 3D Vision!"
Excited (again! 🤓) to share new work from our group, led by @jiahan_zhang1: Generative Cinematographer.
Starting from a single image, artists can plan a shot in 3D by authoring camera moves and choreographing the subjects, then see those choices come to life in the generated video.
Paper: arxiv.org/abs/2610.02180
Project: generative-cinematographer.g…
Was wonderful to watch this work evolve up-close @ToyotaResearch
A paradox in robot learning:
robot policies are spatially dumb and data-inefficient but their image features are spatially robust and semantically rich.
Image encoders are spatially smart — features embed semantics + geometry + are multiview-consistent ...
but robot policies we build upon them are spatially dumb — hundreds of demos to learn simple pick+place and a small camera bump breaks them
To better bridge this gap, check out BIND: a new action head that binds each candidate robot action to its projecting 2D image feature(s)
BIND basically offers the network the info of '‘choosing this candidate robot action would move the robot EEF to this image feature."
Result: policies much more data efficient and robust to OOD viewpoints and OOD object positions (same rgb input, no 3D sensors, dense EEF trajectories out) 🧵
Krishna Murthy retweeted
A paradox in robot learning:
robot policies are spatially dumb and data-inefficient but their image features are spatially robust and semantically rich.
Image encoders are spatially smart — features embed semantics + geometry + are multiview-consistent ...
but robot policies we build upon them are spatially dumb — hundreds of demos to learn simple pick+place and a small camera bump breaks them
To better bridge this gap, check out BIND: a new action head that binds each candidate robot action to its projecting 2D image feature(s)
BIND basically offers the network the info of '‘choosing this candidate robot action would move the robot EEF to this image feature."
Result: policies much more data efficient and robust to OOD viewpoints and OOD object positions (same rgb input, no 3D sensors, dense EEF trajectories out) 🧵
Krishna Murthy retweeted
🙃 you can just do things that look like tactile sensing, without tactile sensors 🙃
💬 Visit my poster at the #IROS2026 workshop Enhancing Dexterous Manipulation Through Haptic Feedback!
📍 Rooms 403/404
🕐 11:40 AM – 12:15 PM
🗓️ Thursday, October 1, 2026 (tomorrow)
Come learn about Blind Dexterity 😎 and how it spawned The Blind Skateboarder 🛹 – at Pittsburgh, PA
In my latest PhD paper, we declare WAR on sensor-maxxing.
For the first time, push-resilient humanoid walking, just with joint encoders! No IMU, no F/T sensors.
🥁 Introducing Blind Dexterity 🧵 👇
w/ @OKaidanov @liu_puze @Jan_R_Peters at @DFKI @ias_tudarmstadt
Krishna Murthy retweeted
1/5 📢 Excited to share PowerSim, which lets us (a) interact with objects in captured scenes, (b) simulate how they move and deform, and (c) render their changing appearance.
This is one of the first papers from the first cohort of PhD students in my new 3P Vision Group at Hopkins 🥹
Demos: power-sim.github.io/
Paper: arxiv.org/abs/2609.38153
Krishna Murthy retweeted
A very interesting experiment is being tried at @RoboticsSciSys
Two stage submissions. (similar to ISRR from 10+ years ago)
Stage 1: 6-page Extended Abstract
Authors submit a 6-page Extended Abstract consisting of up to 5 pages of content and 1 page of references
Complete experiments not expected. (no links or supplements)
Stage 2: Full paper w/ background and experiments.
If experiments yield negative results, that is not a criterion for rejection.
Signed Reviews
Reviewer and Chair names will be made public for papers that are accepted. We believe disclosing the identities of Program Committee members at this final stage will strengthen community bonds between authors and their respective reviewers and chairs.
Signed reviews is a rather big commitment.
Although disclosure may be create unintentional dynamics.
roboticsconference.org/infor…
Attending IROS in Pittsburgh, but only around on Sunday for the workshops. Reach out to say hi / chat!
Krishna Murthy retweeted
Principia will appear at #NeurIPS2026 with three 5s, though it wasn’t selected for a Spotlight or Oral. Reminds me of “Shadows Don’t Lie and Lines Can’t Bend” at #CVPR2024, which had similar faith; still one of my favorite papers.
I would have loved to see the students get that recognition. Still very proud of this work and the effort that went into it.
Huge congrats to @varunvarmat and @shivam_tr! Collecting this dataset required a great deal of patience: setting up calibrated experiments, checking physical properties, and even machining components to ensure they are comparable. It’s hard to convey how much work goes into getting these details right.
With Principia, we are asking a simple question: if a video generator has better physics, can we show it in simple, controlled settings? Relative evaluation with objective metrics gives us a way to test those claims. I’d love to see future models evaluated this way.
@arena, would you be interested in helping us set this up for future video generators?
1/ What do video models know about high school physics? Less than you'd think.
Led by @varunvarmat and @shivam_tr with @rvbabuiisc, we built Principia: relational physics tests for video models. Eight laws, 500+ real scenes, one simple idea.
principiabench.github.io
paper: arxiv.org/abs/2609.04200
Krishna Murthy retweeted
We're hosting the Workshop on Scalable Tactile Sensing for Dexterity at IROS 2026 in Pittsburgh this Sunday.
Join us for fun discussions in this exciting space. There will be talks, posters, and a spicy panel discussion!
tactile-dexterity-iros.githu…
Krishna Murthy retweeted
A new negative-result investigation: closer-to-target (non-Gaussian) action priors did not improve the fine-tuning of Large Behavior Models (LBMs), i.e., robot policies pretrained on large, diverse datasets.
Prior work found that replacing the Gaussian prior of diffusion and flow-matching policies with a closer-to-target prior substantially improves policies trained from scratch. We hypothesized that this benefit would carry over to fine-tuning pretrained LBMs.
Across 100K+ simulation rollouts and 1250 hardware rollouts, we found that it does not.
Website: cxu-tri.github.io/non_gaussi…
Paper: arxiv.org/abs/2609.27070
🧵 (1/10)
Krishna Murthy retweeted
Wrote done my thoughts on what's fundamentally new for vision when using agents --- obviously, apart from some hard problems that have now been nearly solved :)
Krishna Murthy retweeted
📢 We're hiring (interns)!
The Seattle robotics lab @NVIDIARobotics is looking for talented PhD research interns to join us in 2027.
Interns at SRL will have the chance (depending on their preference) to either contribute to one of our ongoing group efforts or work on independent projects/topics of interest on topics including robot foundation models, agentic robotics, and precise manipulation.
Reach out to me if interested in working on robot foundation models 🤖🤖
nvidia.eightfold.ai/careers/…
