Robots can be constructed with fewer resources and less strict constraints than is generally believed. Soft robots can be constructed with very few parts and from a wide variety of materials. This makes them a potentially appealing choice for applications where there are resource constraints on system fabrication. However, soft robot dynamics are difficult to accurately model analytically, due to a multiphysics coupling between shape, forces, temperature, and history of motion. Data-driven methods have successfully captured unmodeled behavior of soft systems on short timescales, but begin to fail during lifetime operation due to nonstationarity: changes in the dynamics over time.
In this work, we propose a memory-based framework for long-term control of soft robots with nonstationary dynamics. We characterize the nonstationary behavior present in existing experimental platforms, and explore extensions to traditional planning and control strategies for the nonstationarity problem. We test these extensions in simulation, and identify multiple hardware platforms suitable for testing, verifying and developing the approach further.
Carmel Majidi (Chair)
Josh Bongard (University of Vermont)
Zoom Participation. See announcement.