Caitlin Freeman, PhD

Caitlin Freeman, PhD

Robotics Researcher and Mechanical Engineer

Research

Data-Driven Gait Synthesis

How do we know what control inputs to use to get a robot to successfully move around in a given environment? For traditional rigid robots, we use established kinematics and dynamics equations to apply model-based control. But this isn’t so simple for soft robots, whose sensivity to manufacturing innacuries and environmental forces alongisde complicated dynamics hinder the feasibility and accuracy of traditional approaches. Instead, we can use a reinforcement learning-based approach, where we collect data on repeated robot motions which are then fed into an optimization algorithm. The result is optimal translation and rotation gaits that are found without using an explicit model of the robot. We can even use these gaits to achieve real-time path planning with obstacle avoidance.

Design Optimization of Modular Reconfigurable Soft Robots

Increasing the functionality, adpatability, and versaility of robots usually means increasing the cost – think of the complex and expensive robots advertised in the news. Alternatively, we can use innovative methods to design and build low-cost robots capable of exploring complex environments. In this research, we focus on two solutions: soft robots and modular reconfigurable robots. Soft robots are able to passively adapt to environmental forces in a way that simplifies control, reduces the required number of actuators, and enables relatively simple designs to exhibit complex behavior. Then, we can build a collection (or swarm) of many soft robots designed to both function individually and combine together into larger robots, which we refer to as modular reconfigurable robots. Figuring out how to design and control these systems ins’t trivial. Nevertheless, we can use modern CAD software and map projection formulas to optimize the design of the robots such that multiple robots can both move individually and combine together to form a ball.