Search papers, labs, and topics across Lattice.
DOT-Sim is introduced as a differentiable simulator for optical tactile sensors, modeling the sensor's physical behavior with the Material Point Method (MPM) to capture large, non-linear deformations. It learns a residual image relative to the real-world idle state to simulate optical responses, enabling rapid calibration using only a few real-world demonstrations. The simulator's realism is validated through zero-shot sim-to-real transfer, achieving high classification accuracy and precise trajectory following in real-world tasks.
Training tactile perception in simulation just got real: DOT-Sim enables zero-shot sim-to-real transfer with high accuracy by precisely calibrating physical and optical sensor properties.
Simulating optical tactile sensors presents significant challenges due to their high deformability and intricate optical properties. To address these issues and enable a physically accurate simulation, we propose DOT-Sim: Differentiable Optical Tactile Simulation. Unlike prior simulators that rely on simplified models of deformable sensors, DOT-Sim accurately captures the physical behavior of soft sensors by modeling them as elastic materials using the Material Point Method (MPM). DOT-Sim enables rapid calibration of optical tactile sensor simulation using a small number of demonstrations within minutes, which is substantially faster than existing methods. Compared to current baselines, our approach supports much larger and non-linear deformations. To handle the optical aspect, we propose a novel approach to simulating optical responses by learning a residual image relative to the real-world idle state. We validate the physical and visual realism of our method through a series of zero-shot sim-to-real tasks. Our experiments show that DOT-Sim (1) accurately replicates the physical dynamics of a DenseTact optical tactile sensor in reality, (2) generates realistic optical outputs in contact-rich scenarios, (3) enables direct deployment of simulation-trained classifiers in the real world, achieving 85% classification accuracy on challenging objects and 90% accuracy in embedded tumor-type detection, and (4) allows precise trajectory following with a policy trained from demonstrations in simulation, with an average error of less than 0.9 mm.
