Skip to main content
GameDev.net gamedev.net

Sim To Real News

The latest Sim To Real coverage curated for game developers.

The Domain Is a Residue: Adapting Self-Supervised Features, Not Generators

arXiv cs.GR details a 2.9M-parameter adapter that edits self-supervised features instead of generators for weather and sim-to-real translation. The Representation Feature Adapter keeps scene structure while moving the domain residue, …

Research arXiv cs.GR · 1 week, 6 days ago
Counterfactual Video Generation Enables Scalable Humanoid Loco-Manipulation

arXiv cs.GR details PRISM, a real-to-sim-to-real pipeline that turns a few human interaction clips into hundreds of training examples for humanoid loco-manipulation. The system uses counterfactual video generation plus contact-anchored …

Research arXiv cs.GR · 1 week, 6 days ago
R2S-EGO: Dual-Proxy Refinement for Sparse-Capture Real-to-Sim

A new real-to-sim pipeline, R2S-EGO, uses dual proxies to patch sparse real-world capture with simulator-generated views. In tests on three Replica scenes and 48 frozen Unitree G1 ego views, six-view …

Research arXiv cs.GR · 2 months ago
Meshless Domain Randomization via Explicit Parameter Perturbation of 3D Gaussian Splatting

Researchers have proposed a meshless domain-randomization pipeline built on 3D Gaussian Splatting, aimed at closing the sim-to-real gap for complex organic assets. Instead of extracting difficult textured meshes, the method …

Research arXiv cs.GR · 2 months, 2 weeks ago
Real2Sim2Real for Vision-Language-Action Manipulation: An AMD ROCm-Based Pipeline

AMD is pushing a full ROCm-based stack for embodied AI, showing that vision-language-action manipulation can run without CUDA lock-in. The demos span sim-to-real robot control, language-grounded object selection, Real2Sim data …

Research arXiv cs.GR · 2 months, 2 weeks ago
NeuralActuator: Neural Actuation Modeling for Robot Dynamics and External Force Perception

NeuralActuator models servo behavior beyond the usual current-to-torque shortcut, aiming to cut sim-to-real drift on low-cost robots. It also infers external forces without dedicated sensors and flags mechanically restricted motion. …

Research arXiv cs.GR · 2 months, 4 weeks ago
VLK: Learning Humanoid Loco-Manipulation from Synthetic Interactions in Reconstructed Scenes

Researchers synthesized 48,000 vision-language-kinematics trajectories from reconstructed indoor scenes to train a humanoid policy without any human demonstration data. The interesting part for developers is the sim-to-real pipeline: 3D Gaussian …

Research arXiv cs.GR · 3 months, 1 week ago
TacCoRL: Integrating Tactile Feedback into VLA via Simulation

TacCoRL reports a 72.5% average success rate on four contact-heavy bimanual tasks by adding tactile feedback to VLA policies and training in simulation. The practical takeaway is that the model …

Research arXiv cs.GR · 4 months ago