Selected projects
Development and mechanical analysis of soft sensors
Using a knitting machine, it is possible to manufacture soft sensors with repeatable behavior from conductive yarn. The sensors can be oriented in different directions (vertically, horizontally, angled) using stitch geometry. Sensor performance can then be analyzed with a combination of mechanical testing and resistance measurements. This work is ongoing.
Reducing energy cost of transport for a direct-drive legged robot on sand
Legged robots are generally able to locomote in the desert, even being able to climb steep dunes for short periods of time. However, they quickly overheat, which can cause damage to the motors. Direct-drive (no gearbox) robots are especially quick to overheat, but represent an attractive architecture for field work because their legs can be used to sense ground properties such as shear strength.
For Sonia’s dissertation, she developed a reactive controller to reduce the energy cost of a direct-drive legged robot running on sand. A reactive controller “reacts” to the error between its current state (position and velocity of joints) and its commanded state, and uses a simple control mechanism such as proportional-derivative control (exerting a spring-like force) to attempt to return to its commanded state. Because desert sand is highly unpredictable and can change stiffness drastically within one robot body length, it is important to use a control mechanism that does not rely on a complex model of the ground which requires data to produce.
Developing a legged pack robot as a field assistant for geoscience research
Geoscientists currently study erosion and other natural processes in deserts in one of two ways. First, they can use arrays of stationary sensors, which enable them to collect a lot of data in one location over a period of time. Second, they can collect samples by hand during field trips, which allows them to collect data over distance but not over time.
An autonomous legged robot which can carry an array of sensors could help geoscientists collect detailed data in both distance and time. It could even be triggered to perform different data collection protocols based on the time of year or the weather reported by nearby stationary sensors.
In this project, we outfitted a robust and well studied legged robot with typical sensors used by geoscientists and a new sensor that uses a single direct-drive leg to measure the force required to scrape off the top layer of grains — a proxy for the erodibility of the ground. We also studied how geoscientists collect data and would collaborate with a legged robot field assistant. This project was led by Prof. Feifei Qian and Dr. Cristina Wilson.
Theoretical foundations of robot behavior design
An affordance is an opportunity for purposeful action for an agent in its environment. We argue that we can — and should — design robots to exploit affordances efficiently. Complex behaviors can be as explainable and robust as simple behaviors if they are built using compositions of simple component behaviors, for example using the tools of dynamical systems theory. Sonia started this project as a collaboration with Prof. Lisa Miracchi and ongoing work is a collaboration with both Prof. Miracchi and Dr. Ben Baker.