Siddharth Ancha
@siddancha
Research Scientist @MIT_CSAIL @MIT | PhD from @SCSatCMU @CarnegieMellon | I work on Robotics 🤖
Diffusion/flow policies 🤖 sample a “trajectory of trajectories” — a diffusion/flow trajectory of action trajectories. Seems wasteful? Presenting Streaming Flow Policy that simplifies and speeds up diffusion/flow policies by treating action trajectories as flow trajectories! 🌐…
Shameless plug: you can get continuous-time robot policies by treating action trajectories as (continuous-time) flow matching trajectories! x.com/siddancha/stat…
Diffusion/flow policies 🤖 sample a “trajectory of trajectories” — a diffusion/flow trajectory of action trajectories. Seems wasteful? Presenting Streaming Flow Policy that simplifies and speeds up diffusion/flow policies by treating action trajectories as flow trajectories! 🌐…
Really nice blog post! I wish we would all blog about the stories behind our papers — our motivations, inception of the idea, trials and tribulations of the project, and even how we decided on our method’s name 🙂. Makes research so much more fun!
This was an incredibly important project to me - I’ve wanted to solve it for years, but had no idea how. This was all @sukjun_hwang and @fluorane's amazing work! I wrote about the story of its development, and what might be coming next. The H-Net: goombalab.github.io/blog/2025/hnet…
This is a really neat idea to improve flow matching! The key observation is that we can analytically compute the marginal velocity field at the boundaries t=0, 1. This lets us design our NN architecture so that it matches the known boundary velocities by construction! @lqiang67
Improving Rectified Flow with Boundary Conditions "Rectified Flow offers a simple and effective approach to high-quality generative modeling by learning a velocity field. However, we identify a limitation in directly modeling the velocity with an unconstrained neural network:…
Hadn't realized that @NVIDIARobotics' Newton simulator (alpha release) is now available open-source on GitHub! github.com/newton-physics… Was super easy to install. This is one of their basic examples, with a somewhat janky viz. Looking forward to see how this matures!
Join experts Yuval Tassa from @GoogleDeepMind and Miles Macklin from NVIDIA at #GTCParis to learn about Newton, an open-source, extensible physics engine for robotics simulation, co-developed by Disney Research, Google DeepMind and NVIDIA. Gain insights into breakthroughs in…
Excited to be speaking tomorrow at the Point Cloud Tutorial at #CVPR2025! We’re diving into all things 3D — from fundamental research in point clouds to industrial applications in Physical AI. 📍 Room 202A 🗓️ June 11, 3pm CDT Thanks @KaichunMo @XiaoyangWu_ for the invite!
Join the 2nd Point Cloud Tutorial for #CVPR2025 Theme: All You Need to Know About 3D Point Cloud Date: June 11, full-day length Location: 202A TL;DR: For the 2nd point cloud tutorial at CVPR 2025, we aim to move beyond traditional topics like backbone design and…
🎉🎉 Our competition proposal was accepted to @NeurIPSConf 2025!! Excited about this competition not only as a technical challenge but also for the nostalgia! 👀Stay tuned for more details
Really excited to share that our Pokémon competition is accepted at NeurIPS 2025!🔥 Get ready for an epic AI challenge featuring both speedruns🎮 and multi-player Pokémon battles⚔️---more details coming soon!
Cool work on flow matching to separate a mixture of k sources, by defining a permutation-equivariant flow in the (k-1) dimensional orthogonal subspace! With neat audio separation results!
Finally wrapped up my first work after joining @GoogleDeepMind last year 🥳 FLOSS: Flow matching meets Source Separation! with John Hershey, @ArnaudDoucet1 , and Henry Li. 📜 arxiv.org/abs/2505.16119 🔊 google.github.io/df-conformer/f…
Check out "Streaming Flow Policy" at #ICRA2025! We simplify diffusion/flow policies by treating action trajectories as flow trajectories! ▶ Late breaking poster session (Thu 9.30am, Hall A1/A2, ThLB1R.4) ▶ Spotlight talk + Poster at ICRA Beyond Pick and Place Workshop (Fri…
Diffusion/flow policies 🤖 sample a “trajectory of trajectories” — a diffusion/flow trajectory of action trajectories. Seems wasteful? Presenting Streaming Flow Policy that simplifies and speeds up diffusion/flow policies by treating action trajectories as flow trajectories! 🌐…