Associate Prof. @Cambridge_Uni in ML/AI + Robotics | earlier: @Stanford Postdoc, @kth_rpl PhD, @CMU_Robotics MS, @Google engineer (search, OCR+Books/StreetView)
Cambridge, UK
Joined December 2009
- Tweets222
- Following911
- Followers2.5K
- Likes2.5K
From aerodynamic wing design to robot fingertips, Structural Affinity is our new method for eliminating jagged 3D deformations at the parameterization level, producing smoother, manufacturable shapes.
Avi @projectavi_n covers the specifics in a visually engaging thread.
1/ Announcing Structural Affinity: smooth, plausible cage-based 3D deformations, no extra loss terms. Move one handle and its structural neighbours follow. Tested on neural cages, wings, and robots.
With Shengan Yang, Hantao Zhong and Rika Antonova.
camral.github.io/structural-… 🧵
It was great to talk about robot evolution and codesign at the Robot Talk Podcast last week! Curious to hear the feedback from the community, especially now that we are stepping up the efforts to incorporate more environments into our Co-design Gym
Full "Evolving robot bodies and brains" podcast episode: robottalk.org/2026/10/16/epi…
We brought KernelCraft to #ICML2026 to show that LLM agents can now tackle kernel engineering, translating AI models into optimized code for specialized physical hardware, and eliminating the time-consuming aspects of manual low-level code design👇🧵 1/3
KernelCraft kernelcraft-cam.github.io shows how to use agentic systems to generate close-to-metal assembly for custom hardware, spanning commercial accelerators like Coral and AMD NPUs, and emerging AI accelerator designs, e.g. open-source PLENA, which we just presented at ISCA2026
KernelCraft & PLENA were led by PhD students Jiayi Nie & George Wu from @Cambridge_Uni, as part of the larger AIxSim @ARIA_research project with @imperialcollege & @EdinburghUni, with a goal to enable simulation & co-design for novel accelerator hardware: aicrosssim.github.io/researc…
We will present our work on differentiable tactile simulation at #ICRA2026 next month. This has been a fruitful collaboration with Yang You, Won Kyung Do, Aiden Swann, Monroe Kennedy III, and Leonidas Guibas from @Stanford.
Looking forward to comments further collaborations!
🚀 Excited to announce that DOT-Sim is accepted to #ICRA2026! We build a differentiable optical tactile simulator that captures both the physics and optics of soft sensors: an MPM-based elastic model handles large, non-linear deformations, and a learned residual renderer maps depth+normal to realistic sensor images. It calibrates to a real DenseTact in minutes, and enables zero-shot sim-to-real on object classification (85%), embedded tumor detection (90%), and sub-mm trajectory following. #ICRA2026 #Robotics #TactileSensing #Simulation #Sim2Real #AI
Joint work with amazing collaborators: Won Kyung Do, Aiden Swann, Rika Antonova, Monroe Kennedy, and Leonidas Guibas.
🔗 ArXiv: arxiv.org/abs/2604.27367
The ICRA workshop on dexterous manipulation is accepting submissions now (until April 10th). Dexterous manipulation is still very challenging, but is already at the stage where rapid progress can be made. So, the workshop should be quite interesting this year!
🤖 Attending ICRA 2026 and interested in dexterous manipulation? Don’t miss the chance to join or submit to our workshop!
📷Checkout our outstanding lineup of speakers, best poster award, and more details on our website: shorturl.at/ojsTo.
Deadline: April 10 (AOE).
Join an exciting project @Cambridge_Uni on co-design of hardware & control for dexterous mobile manipulation. Looking for roboticists with industry or practical academic experience. Can be a postdoc option, but MS/BS graduates are welcome too.
Interested? Please message/email me
Rika Antonova retweeted
Today we’re excited to unveil a new generation of Segment Anything Models:
1️⃣ SAM 3 enables detecting, segmenting and tracking of objects across images and videos, now with short text phrases and exemplar prompts.
🔗 Learn more about SAM 3: go.meta.me/591040
2️⃣ SAM 3D brings the model collection into the 3rd dimension to enable precise reconstruction of 3D objects and people from a single 2D image.
🔗 Learn more about SAM 3D: go.meta.me/305985
These models offer innovative capabilities and unique tools for developers and researchers to create, experiment and uplevel media workflows.
Mobi-π @ CoRL today: come to our poster to see how manipulation policies trained with stationary (fixed-base) robot arms can be mobilized for versatile mobile manipulation.
We are presenting Mobi-π today at CoRL 2025.
Spotlight: Session 5, 1:30pm
Poster: #28, 5-6:30pm
Come chat with us and learn more about our work!
Rika Antonova retweeted
We present CUPID 💘 at @corl_conf today.
Spotlight: 15:30-16:30
Poster: 16:30-18:00, Poster #44
CUPID 💘 lets us trace back policy performance to training demonstrations using influence functions.
This helps with curating demonstrations for DPs and VLAs.
#CoRL2025 presentation coming up on September 28th: stop by our poster to learn more about CUPID - a robot data curation method based on a novel influence function-theoretic formulation for imitation learning policies (spotlight session 2, poster session 1).
Only a few hours until #CoRL2025! Before that, let's look at the "Optimistic Future of Robotics Industry" 2024 panel, to see which milestones can be marked as done, and which ones need the new academic and industrial research advances from 2025 and onwards youtube.com/watch?v=iZwhOZej…
Even simple-looking puzzles can reveal the limits of human physical reasoning. Can AI do better? Stop by to chat with us at ICML to find out (we will present our work on causality-based physical reasoning this Thursday).
Rika Antonova retweeted
What makes data “good” for robot learning? We argue: it’s the data that drives closed-loop policy success!
Introducing CUPID 💘, a method that curates demonstrations not by "quality" or appearance, but by how they influence policy behavior, using influence functions. (1/6)
Enabling mobile manipulation for long-horizon multi-phase tasks with HoMeR -- a whole-body control method that learns when to use long-range actions versus when to switch to fine-grained precise manipulation.
Reinforcement Learning pioneers got a Turing award! Their enthusiasm helped RL survive 'winters', inspired several generations of RL researchers, and ensured focus on far-reaching research questions. It has been amazing to be a part of this community.
Meet the recipients of the 2024 ACM A.M. Turing Award, Andrew G. Barto and Richard S. Sutton! They are recognized for developing the conceptual and algorithmic foundations of reinforcement learning. Please join us in congratulating the two recipients! bit.ly/4hpdsbD
