26 September 2024
As we strive to integrate robotics into human-centric environments, guaranteeing safety becomes an absolute priority. While safety is traditionally ensured through computational control policies, this approach is vulnerable to perception errors and often results in overly cautious behavior that limits robot performance. Soft robotics presents a promising alternative by establishing passive compliance throughout the entire robot body with material softness. However, the modeling and control of continuum soft robots presents significant challenges due to their infinite degrees of freedom, complex nonlinear dynamics, and time-dependent behaviors such as hysteresis.
In this work, EMERGE partners from the Delft University of Technology propose that integrating learned models with model-based controllers presents a compelling alternative approach, combining the strengths of both existing methods: data-driven models that demand less expert knowledge and controllers with interpretable, provably stable behavior. The central challenge lies in determining the essential characteristics and structures a learned model must exhibit to facilitate the application of established model-based control strategies, such as PID-like feedback with feedforward control while ensuring the closed-loop robot system remains compliant and safe.
Read the paper in the link below.
More Information
