PG 2025 (Journal Track) [PDF] [Project] [Code]
We address the challenges of existing mesh simplification methods that cause self-intersections and surface distortions. Our GPU-based approach performs parallel re-meshing, intersection-free simplification, and safe projection to produce clean, geometry-preserving meshes suitable for efficient rendering, editing, and simulation.
SIGGRAPH 2025 (Journal Track) [PDF] [Code]
Continuous Collision Detection (CCD) in IPC-based simulations is slow and often requires powerful GPUs. We introduce a sequential CCD algorithm for convex shapes with affine trajectories (as in ABD) that achieves a 10x speed-up over traditional primitive-level CCD. Our method uses cone casting, a generalization to the ray-casting CCD used in traditional physics engines for rigid bodies.
Manipulating deformable objects like cloth is difficult due to their complex dynamics and tricky state estimation. We propose a generative, transformer-based diffusion model that handles both perception and dynamics. Our method reconstructs the full cloth state from sparse observations and predicts future movement, reducing long-horizon prediction errors by an order of magnitude compared to previous methods. This framework successfully enabled a real robot to perform complex cloth folding tasks.
We propose a learning framework and system that automatically decomposes task demonstrations into semantically meaningful skills using off-the-shelf foundation models, and generates diverse synthetic demonstration datasets from a few human demos through reinforcement learning. These sim-augmented datasets enable robust skill training, with a Skill Routing Transformer (SRT) policy effectively chaining the learned skills together to execute complex long-horizon manipulation tasks.
TRO 2024 [PDF]
We build a general-purpose Sim2Real protocol for manipulation policy learning with marker-based visuotactile sensors. To improve the simulation fidelity, we employ an FEM-based physics simulator that can simulate the sensor deformation accurately and stably for arbitrary geometries. We further propose a novel tactile feature extraction network that directly processes the set of pixel coordinates of tactile sensor markers and a self-supervised pre-training strategy to improve the efficiency and generalizability of RL policies.
