SoftMAC: Differentiable Soft Body Simulation with Forecast-based Contact Model and Two-way Coupling with Articulated Rigid Bodies and Clothes

IROS 2024

1Carnegie Mellon University, 2National University of Singapore

Highlights

  • We propose a novel forecast-based contact model for MPM, which reduces penetration without introducing artifacts like unnatural rebound.
  • We present a penetration tracing algorithm for the contact between MPM and non-volumetric cloth meshes.
  • To the best of our knowledge, SoftMAC is the first differentiable robotic simulator to support two-way dynamics soft-rigid and soft-cloth coupling.

Abstract

Differentiable physics simulation provides an avenue to tackle previously intractable challenges through gradient-based optimization, thereby greatly improving the efficiency of solving robotics-related problems. To apply differentiable simulation in diverse robotic manipulation scenarios, a key challenge is to integrate various materials in a unified framework. We present SoftMAC, a differentiable simulation framework that couples soft bodies with articulated rigid bodies and clothes. SoftMAC simulates soft bodies with the continuum-mechanics-based Material Point Method (MPM). We provide a novel forecast-based contact model for MPM, which effectively reduces penetration without introducing other artifacts like unnatural rebound. To couple MPM particles with deformable and non-volumetric clothes meshes, we also propose a penetration tracing algorithm that reconstructs the signed distance field in local area. Diverging from previous works, SoftMAC simulates the complete dynamics of each modality and incorporates them into a cohesive system with an explicit and differentiable coupling mechanism. The feature empowers SoftMAC to handle a broader spectrum of interactions, such as soft bodies serving as manipulators and engaging with underactuated systems. We conducted comprehensive experiments to validate the effectiveness and accuracy of the proposed differentiable pipeline in downstream robotic manipulation applications.


Contact Model

To alleviate penetration, we introduce a novel method called forecast-based contact model to manage the boundary conditions for Material Point Method (MPM). Specifically, the model takes a grid-to-particle transfer to look ahead in the grid operation stage, imposes constraints on particles within the contact region, and then adjusts the grid velocity accordingly.
Forecast-based contact model requires signed distance fields (SDFs) to penalize penetration. While the definition of SDF is straightforward for volumetric objects, it is hard to determine the sign on nonvolumetric meshes. To solve the problem, we propose a penetration tracing algorithm that capitalizes on the localized motion of particles to reconstruct the SDF within confined zones. In this way, the contact model can be applied to both soft-rigid and soft-cloth coupling.

fail fail fail
Figure 2. Comparisons between three different contact models. In this experiment we pour water into a thin glass. Grid-based model (left) leads to severe penetration. Particle-based model (middle) also causes a few particles to penetrate the glass. Our forecast-based model (right) achieves the most robust performance.

fail fail
Figure 3. Effectiveness of penetration tracing. Drag four corners of a towel to squash a plasticine. Towel goes through the plasticine without penetration tracing (left). Both the plasticine and towel deform due to the contact after adding the algorithm (right).

Trajectory Optimization

We demonstrate the effectiveness of SoftMAC across six robotic manipulation tasks. For each task, we optimize an action sequence to minimize the task-specific loss using the gradient information calculated by SoftMAC. The initial, intermediate, and final trajectories are displayed from left to right.

fail
(a) Pour wine (rigid2mpm)
fail
(b) Pour wine with Franka (rigid2mpm)
fail
(c) Squeeze plasticine (rigid2mpm)
fail
(d) Pull door (mpm2rigid)
fail
(e) Make taco (cloth2mpm)
fail
(f) Push towel (mpm2cloth)

fail
Figure 4. Experiment results of coupled differentiable simulation. (a) Control glass to pour the liquid into a bowl. (b) Control Franka arm to pour the liquid into a tank from a bottle. (c) Control Franka gripper to squeeze the plasticine into target shape. (d) Control 2 selected points on the tortilla to fold the taco into target shape. (e) Control soft gripper to pull the door into target angle. (f) Control soft gripper to push the towel into target pose. The training loss curves are plotted below.

Comparison with Other Simulators

fail
Table 1. Comparison with other popular MPM-based simulators. To the best of our knowledge, SoftMAC is the first differentiable robotic simulator to support two-way dynamics soft-rigid and soft-cloth coupling.

BibTeX

@inproceedings{liu2023softmac,
  title={SoftMAC: Differentiable Soft Body Simulation with Forecast-based Contact Model and Two-way Coupling with Articulated Rigid Bodies and Clothes},
  author={Liu, Min and Yang, Gang and Luo, Siyuan and Shao, Lin},
  booktitle={IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS) },
  year={2024},
}