Different from motion planning and tracking, I-BFM uses FB representation to handle the box-carrying task. In the demonstration, the robot performs the pick-and-place task without palpable phase switching. FB makes the whole task into a "higher-level optimal control problem" that aims to move the box to the goal in any way with the lowest "energy".
3DGS looks real—until you shift the camera or insert a car. 🚗
Excited to unveil SPVC, our joint work with @GWMGlobal!
4 upgrades to photorealistic driving video fixing:
🔹 Structured: Geometry-guided, not unconstrained.
🔹 Panoptic: Background + foreground, not just one.
🔹 Video: Temporally consistent, not frame-by-frame.
🔹 Cross-Dataset: One model across Waymo, nuScenes & PandaSet.
One unified video diffusion fixer. See the difference! 👇
📄 Paper: arxiv.org/pdf/2608.17420
🌐 Project: li00147.github.io/SPVC-Project-P…
Our manipulation results were mind-blowing the first time I saw them, but earlier on, my favorite result was this locomotion run. Many great people have done parkour-type skills with locomotion already, but as far as I am aware, most, if not all, of these works trained separate
Our manipulation results were mind-blowing the first time I saw them, but earlier on, my favorite result was this locomotion run. Many great people have done parkour-type skills with locomotion already, but as far as I am aware, most, if not all, of these works trained separate policies for each terrain, and then distilled into a single policy. They also typically contained tons of hand-engineered rewards and curriculums.
A very early motivation for SGS was to simply figure out if we could get similar results but get rid of as much of this hand-engineering as possible, and we managed to achieve that! This video shows a continuous 1-min run in much harder terrains than ever before, all solved by the same policy.
Can humanoid motion generation and control be unified in a single policy?
PredActor uses a single end-to-end diffusion policy rather than separate motion generation and tracking. With proprioceptive feedback, it predicts internal future states and directly outputs actions at 50 Hz onboard a Unitree G1. Text-conditioned motion, joystick steering, disturbance response, and semantic interpolation are supported in the single policy.
Project Page: masteryip.github.io/predactor.gith…
@ChenTessler many thanks for your feed back!
I will refine the positioning of our data collection and training pipeline more precisely as soon as possible!
More Demonstration in MuJoCo Simulator, including Target tracking, Text Control, Text-Joy Combined Control, Disturbance.
Mujoco evaluation codes can be found at the Repo: github.com/MasterYip/Pred…
Unlike a motion-generator-plus-tracker pipeline, it uses proprioceptive feedback to jointly predict executable actions and an internal future-state trajectory. Only actions are executed; predicted states remain inside the policy as a guidance interface. Classifier-free guidance strengthens text-conditioned behaviors, while classifier guidance steers predicted states toward test-time objectives. Everything can be fully executed on Jetson Orin NX at 50 Hz.
518 Followers 4K FollowingLLM Post-Training/ Recursive Self-Improvement for Reasoning
Co-founder/ MTS @ReasonCoreAI
PhD @UCLA, Bachelors @iitbombay
Led a team of ~500 PhD researchers
47 Followers 149 FollowingHaiyun Zhang is a Ph.D. in @texas_robotics at @UTAustin
. X posts his engineering projects backup. Ongoing works may not be posted, but feel free to DM.
518 Followers 4K FollowingLLM Post-Training/ Recursive Self-Improvement for Reasoning
Co-founder/ MTS @ReasonCoreAI
PhD @UCLA, Bachelors @iitbombay
Led a team of ~500 PhD researchers
6K Followers 1K FollowingAsst. Prof@ Tsinghua University, Co-founder @GalaxeaDynamics. Former Scientist@Waymo, MIT PhD’19. I research on multimodal learning, self-driving and robotics!
1K Followers 169 FollowingCo-founder of Archon @archon_robotics. Assistant Prof @HKUniversity. I direct @OpenDriveLab. We only live once, if we do it right, once is enough.
542 Followers 330 Following🇨🇭phd student @EPFL_en | working on embodied ai and computational neuroscience | previous sr at google, bytedance | open science
11.4M Followers 79 FollowingPublic media for public good. We don't pursue corporate interests, nor yield to pressure of ideological stigmatization and political bias
Our App: https://t.co/uYjoablw4w
963 Followers 678 Following🤖 · 🚵♂️ · 🎸 · 🇪🇨
PhD student @uwcse currently kicking it at @FlexionAI, working on scaling RL. I like biking long distances and playing the bass.
8K Followers 1K FollowingAssistant Professor at @NUScomputing | Robotics & AI PhD @uwcse | Host of @RoboPapers | Ex-@allen_ai robotics lead, @NVIDIA | Magician |
my opinion is my alone.
47 Followers 149 FollowingHaiyun Zhang is a Ph.D. in @texas_robotics at @UTAustin
. X posts his engineering projects backup. Ongoing works may not be posted, but feel free to DM.
12K Followers 781 FollowingProfessor of Computer Vision, @bristolcs. Senior Staff Research Scientist @GoogleDeepMind - passionate about the temporal stream in our lives.
33K Followers 221 FollowingChina EV news, data, insights and more.
Telegram: https://t.co/2fzq9AMRNz
Newsletter: https://t.co/scncnqxkNY
Google News: https://t.co/0Xe3cv6HMP
11K Followers 981 FollowingTweets from Journal of Chemical Information and Modeling (JCIM) and Journal of Chemical Theory and Computation (JCTC) from @ACSPublications
17K Followers 806 Followingdiscovery chemistry nerd @Merckgroup; oncology, immunology, targeted protein degradation, ADCs; soccer, politics, tapas can slip into the feed; my own views