Cheng Lin

Very excited to see recent MBRL/world-model progress. Reported training times went from 72 min for SuperMimic (H100) to 7 min for InstantTrack v1.2 (5090) to 3.4 min for BlinkMimic (4090)! Congrats on those great works! This repo looks at which “implementation factors” help these methods train better. We started it to share implementation tricks, so when people (or coding agents) try to use these models, they can save time and effort. We're adding tasks and baselines, and testing engineering tricks. github.com/Frank-ZY-Dou/ndm-… #Simulation #Robotics #MBRL #BPTT #Humanoid #Graphics

Our work #S4R has been accepted to SIGGRAPH Asia 2026 / ACM Transactions on Graphics (TOG). Removing unintended interpenetration is an important step toward making generated or assembled 3D scenes usable for physical simulation and robotic interaction. S4R tackles this problem through a simple scale-space strategy: we first shrink objects to create clearance, then progressively scale them back to full size while optimizing their poses. This transforms a difficult deep-interpenetration problem into a sequence of mostly shallow-contact quadratic programs (QPs). We are also integrating S4R into an agentic framework for automatically generating and optimizing penetration-free, simulation-ready scene layouts. Project Page: frank-zy-dou.github.io/proje… Code: github.com/Frank-ZY-Dou/Dyna… #Geometry #3D #Graphics #Simulation #Optimization #Robotics

Excited to share our work at #RSS: NeuralActuator: Neural Actuation Modeling for Robot Dynamics and External Force Perception We have now released the complete codebase, datasets, and pretrained checkpoints: 🔗 Code: github.com/Frank-ZY-Dou/Dyna… 📑 Paper: people.csail.mit.edu/frankzy… 🏠 Project Page: people.csail.mit.edu/frankzy… The release includes: * Complete training, evaluation, inference, and rollout-rendering code; * The Neural Actuation Dataset and pretrained checkpoints for OpenManipulator-X, SO-101, and Franka Panda; * Two inference modes: 1. Differentiable-dynamics rollout for predicting robot dynamics; 2. A virtual force sensor that estimates external forces directly from motor telemetry; * Teleoperation and data-collection code, together with detailed hardware setup and component-sourcing instructions; * Fully reproducible evaluation scripts and updated benchmark results. The released checkpoints achieve even better performance than the results reported in the paper; * Additional discussions and experiments on: • Cross-platform scalability; • Direct versus residual torque modeling, with and without model-based inductive bias; • Motor-condition detection. The datasets and pretrained models are also available on Hugging Face. The dataset page additionally includes camera recordings from our data-collection sessions: 📊 Dataset: huggingface.co/datasets/fran… 🤗 Models: huggingface.co/frankzydou/Ne… We hope NeuralActuator can serve as a useful and accessible tool for robot dynamics modeling, differentiable simulation, and external force perception. We are also inviting the community to contribute more diverse actuation data across different robots, actuators, operating conditions, and hardware platforms. We would love to expand the dataset together with the robotics community. We are incredibly honored that NeuralActuator has been selected as a finalist for both the Outstanding Student Paper Award and the Outstanding Paper Award! At this point, I think it is only fair to say: RIP to all the motors we burned along the way🙏 Their sacrifices were not in vain. Please check out our work at #RSS! #Robotics #RobotLearning #DynamicsModeling #ForceSensing #Simulation #RSS #RSS2026