Leveraging Embodied Intelligence for Dexterous Robotic Manipulators Through Iterative Co-design

Abstract The role of embodied intelligence (EI) has the potential to overcome current limitations in the fabrication, control, and resulting behavior to create robust and effective dexterous robotic manipulators. To develop hands that truly exploit EI, we must design hands by considering the entire system: the physical body, sensory systems, and the brain (the controller). However, we lack clear approaches and methods that enable this system level design for hands. We introduce an iterative approach for co-design which seeks to utilize simulation and real world evaluation to maximize the performance by distributing EI across the different elements of the system. To achieve this vision we require hands that can be rapidly fabricated with variability in the design space. Thus, to further the development of robotic hands that utilize EI we need streamlined fabrication pipelines which incorporate spatially distributed sensors, complex geometries and materials, and control distributed at the sensory-motor and high task planning domains.