SoRoMoX: GPU-Accelerated Differentiable Soft Robot Modeling Framework
Researchers have introduced a new modeling framework for soft robots called SoRoMoX, which leverages JAX for efficient GPU-parallel simulations that focus on control and differentiation. This innovative approach includes articulated, Piecewise Constant Strain, and Variable Strain models within a unified interface, providing crucial components like inertia matrices and Jacobians. Notably, it's the first rod/strain-based framework for soft robots that works seamlessly on GPUs and is fully differentiable regarding various parameters. It boasts CPU rollouts that outperform leading alternatives by up to 18.1 times, alongside enhanced GPU capabilities. This development tackles the challenges in soft-robot control, which often falls short of the advanced workflows found in rigid robotics. The framework is built entirely in Python/JAX, and the research can be accessed on arXiv with the identifier 2608.06650.
Key facts
- SoRoMoX is a fully numerical, JIT-compilable Python/JAX framework for soft robot modeling.
- It implements articulated, Piecewise Constant Strain, and Variable Strain models.
- Provides a unified, control-ready interface with inertia matrices, gravitational and elastic forces, Jacobians, and derivatives.
- First rod/strain-based soft-robot modeling framework that runs directly on GPUs.
- End-to-end differentiable with respect to states, inputs, and parameters.
- Sequential CPU rollouts are up to 18.1x faster than state-of-the-art alternatives.
- GPU performance is also improved.
- Paper available on arXiv with identifier 2608.06650.
Entities
Institutions
- arXiv