Collaborative Research: Artificially-Evolved Modular Robotic Swimmers for Enhanced Mobility in Challenging Hydrodynamic Environments

This project will fund research that strives to enable swimming robots with novel capabilities customized to a specified range of objectives and environments. Fish, shaped by hundreds of millions of years of evolution, display a diversity of body structures and neural circuits in response to ecological pressures. This project will build upon modular representations of these evolutionary solutions to implement a robot design process emulating natural selection. The project envisions design features grouped into the following three categories: (i) external flow sensing, decision-making, and power management, (ii) body and fin actuation, shape and internal state sensing, and buoyancy control, and (iii) body and fin shape and compliance control. Automated printing and packaging will allow rapid prototyping of candidate robots from this design space. Each robot will undergo physical tank trials using reinforcement learning to develop control policies for a set of characteristic movements, including speed and acceleration of turning, forward, backward, and sideways motion, and energy efficiency during sustained forward motion, which will then be evaluated by physical flow testing subject to an anticipated range of operating conditions. Candidates will compete against each other to accomplish movement-based tasks in relevant flow conditions, with high-scoring designs selected as the starting point for the next round of testing, and low-scoring designs eliminated from further consideration. After multiple such rounds, the winning configurations will be equipped with fluid flow sensors, gyros, and accelerometers, and will learn decision-making and feedback strategies for choosing and blending individual motion primitives to effectively achieve higher-level guidance and navigation objectives. This work will accelerate the application of intelligent underwater robots to address national needs and grand challenges, including search and rescue, disaster recovery, pollution and ecological monitoring, and infrastructure inspection. Associated outreach and STEM education efforts include developing a plug-and-play robot kit and a science class at the Harvard Museum of Natural History. This research will create modular robotic swimmers capable of artificial evolution, to enable novel swimming capabilities such as stable swimming in turbulent flows, navigation towards wakes of underwater objects, performing stable rheotaxis, and dynamic energy savings via real-time adjustment of robot body and caudal fin stiffness and shape. The project will first modularize fish-inspired robotics to create a Modular, Mutational, Morphing Underwater Robot (M3UBot) design space. Next, asynchronous evolution will be performed directly in the physical M3UBot design space for evolving body morphologies and learning motor control programs for modular swimming behaviors (e.g., rapid turning, acceleration, steering, forward or backward swimming). The large-scale robot evolution in physical space and the “plug-and-play” robot assembly will be enabled by innovating 3D-Printing and Electronic packaging (3DPE) for rapid design and automatic fabrication of M3UBot modules. Finally, selected prototypes from the evolved M3UBot population will be equipped with hydrodynamic pressure sensors; they will then undergo reinforcement learning for feedback control and decision-making that combines modular behaviors to navigate in challenging hydrodynamic conditions. Together, this project will transform the fundamentals and applications of underwater robotics, culminating in next-generation intelligent robotic swimmers capable of hydrodynamic perception, active shape morphing and stiffness tuning, and versatile motor skills in challenging hydrodynamic conditions.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.