Moritz Geilinger,David Hahn,Jonas Zehnder,Moritz Bächer,Bernhard Thomaszewski,Stelian Coros
Abstract:We present a differentiable dynamics solver that is able to handle frictional contact for rigid and deformable objects within a unified framework. Through a principled mollification of normal and tangential contact forces, our method circumvents the main difficulties inherent to the non-smooth nature of frictional contact. We combine this new contact model with fully-implicit time integration to obtain a robust and efficient dynamics solver that is analytically differentiable. In conjunction with adjoint sensitivity analysis, our formulation enables gradient-based optimization with adaptive trade-offs between simulation accuracy and smoothness of objective function landscapes. We thoroughly analyse our approach on a set of simulation examples involving rigid bodies, visco-elastic materials, and coupled multi-body systems. We furthermore showcase applications of our differentiable simulator to parameter estimation for deformable objects, motion planning for robotic manipulation, trajectory optimization for compliant walking robots, as well as efficient self-supervised learning of control policies.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is to develop a differentiable simulator capable of handling the dynamic simulation of rigid bodies and deformable objects with frictional contact. Specifically, the authors propose a new method to deal with frictional contact problems in multi - body systems. These problems are very common in the real world but computationally challenging, especially because they introduce strong nonlinearity and quasi - discontinuities. These problems make it difficult for traditional numerical solvers to handle them efficiently and accurately.
### Main Problems
1. **Non - smooth Nature of Frictional Contact**: Frictional contact usually leads to discontinuities in forces and their derivatives, which poses a significant obstacle to gradient - based optimization (such as inverse simulation).
2. **Coupling of Rigid Bodies and Deformable Objects**: A unified framework is required to handle the dynamics of rigid and soft bodies simultaneously, which were usually treated separately in previous research.
3. **Differentiability Requirements in Inverse Simulation**: In order for the simulation results to be used in tasks such as parameter estimation, motion planning, and self - supervised learning, the simulator must be differentiable and be able to balance between accuracy and the smoothness of the objective function.
### Solutions
To solve the above problems, the authors propose the following methods:
- **Smooth Frictional Contact Model**: Deal with normal and tangential contact forces through the principle - based mollification method, thus avoiding the inherent non - smoothness problem of frictional contact.
- **Implicit Time Integration**: Combine a fully implicit time - integration method to ensure the stability and efficiency of the dynamic solver.
- **Adjoint Sensitivity Analysis**: Use adjoint sensitivity analysis to analytically calculate the derivatives of the simulation results, thereby achieving gradient - based optimization.
### Application Areas
- **Parameter Estimation**: Estimate the stiffness, damping, and frictional characteristics of materials from actual experiments.
- **Robotic Manipulation**: For example, control the manipulation of multi - body systems, including multi - bounce throwing, dragging, and sliding different types of objects.
- **Trajectory Optimization**: Generate optimal motion trajectories for walking robots with compliant actuators and soft feet.
- **Self - supervised Learning**: Embed the simulator in a neural network to learn control strategies in a self - supervised manner, such as a robotic arm learning how to throw a ball to make it land at a specified location.
### Summary
The main contribution of this paper lies in providing a stable, differentiable, two - way - coupled simulation framework suitable for rigid and soft - body systems. Through the smooth frictional contact model, this framework can achieve a good balance between differentiability requirements and the accuracy of modeling real - world behavior. In addition, the authors also demonstrate the potential of this method in multiple application areas, verifying its effectiveness and practicality.