DOT-Sim: Differentiable Optical Tactile Simulation with Precise Real-to-Sim Physical Calibration
Yang You, Won Kyung Do, Aiden Swann, Rika Antonova, Monroe Kennedy, Leonidas Guibas
AI summary
Problem
Simulating optical tactile sensors is difficult due to complex material deformation and internal light transport, causing existing simplified models to fail at accurate sim-to-real transfer.
Approach
The framework combines differentiable Material Point Method simulation for fast physical calibration with a neural network that learns residual optical images relative to a real-world idle frame.
Key results
- Reduces 3D deformation error with F-Score @1mm of 69.89
- Improves optical image PSNR by 17.34% over baselines
- Achieves 85% object classification and 90% tumor detection accuracy in zero-shot sim-to-real transfer
- Enables precise trajectory following with less than 0.9 mm average real-world error
Why it matters
Provides a fast, accurate simulation pipeline that enables reliable training of tactile perception and control policies for soft robots and medical devices.
Abstract
Simulating optical tactile sensors presents signif- icant challenges due to their high deformability and intricate optical properties. To address these issues and enable a physi- cally 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, and (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 policy trained from demonstrations in simulation with an average error of less than 0.9 mm.