Publications
Books
For You, Humans
Mitchell Fogelson, Xinkai Chen, Christopher Dare, Qeifei Dong, Tony Qin
A book of limericks entirely imagined and written by artificial intelligence by retraining GPT-2 on 90K limericks. There are 10 chapters of limericks that we curated for the readers, including topics revolving around Life, Love, Philosophy, and more. The book is full of fun and imaginative imagery that could only be created by a non-human mind. The book enables a beautiful mental process of finding meaning where none may have been intended the way we humans draw these connections. Can AI be useful for human creative thought and creation? What will a book like this look like in just 5 more years with the progress in technology? Can AI create beauty and meaning to supplement the lives of humans?
Journal Publications
GCP-HOLO: Generating High-order linkage graphs for Path synthesis
Mitchell B. Fogelson, Conrad Tucker, Jonathan Cagan
One degree of freedom (1DOF) linkages are persistent in mechanical systems. However, designing linkages to follow a desired path, known as path synthesis, is challenging due to non-linearities, combinatorial nature, and strict geometric constraints. Current state-of-the-art algorithms cannot scale well to linkages with higher-order linkage graphs. One reason for this is that state-of-the-art algorithms spend the majority of the time exploring constraint-violating designs. This work uses an Assur group 0DOF linkage as a graph grammar rule to modify both linkage graph and spatial parameters, ensuring all designs are valid 1DOF linkages. Using this graph grammar, this paper formulates linkage path synthesis as a tree search and uses a Deep Reinforcement Learning (DRL) agent to search the space of kinematically feasible planar 1DOF linkages. This paper introduces a method using a Graph Convolution Policy for High-Order Linkage Graph Optimization called GCP-HOLO. An any-time algorithm, GCP-HOLO outputs linkages with 1-8 loops (4-16 bars) efficiently. When comparing the GCP-HOLO formulation to a recent state-of-the-art paper that solves a Mixed Integer Conic Program, GCP-HOLO generates sets of solutions of varying linkage complexities to 8 test trajectories in a quarter of the time. Extending GCP-HOLO with a global node optimization, such as Covariance Matrix Adaptation Evolutionary Strategy, the results quickly converge to finding better solutions for 4/8 tests, with the whole pipeline capable of a 13X speed increase.
CONfrence Papers
CAPO: Control and Actuator Placement Optimization for Large-Scale Problems with Nonlinear Dynamics
Mitchell B. Fogelson*, Giusy Falcone, Zachary Manchester
(Chicago, IL WAFR 2024)
Actuator selection and placement are critical aspects of robot design that are tightly coupled to task performance. Jointly optimizing actuator placement and control policies or trajectories can enhance efficiency. However, current optimization methods struggle to scale to high-dimensional systems with many degrees of freedom, including soft robots, large flexible spacecraft, or cloth manipulation. We propose CAPO, a scalable Control and Actuator Placement Optimization method, that jointly optimizes the number, type, and placement of actuators, along with control trajectories. CAPO concurrently optimizes over all the control and design parameters in a single computationally tractable nonlinear program that scales favorably with system size and complexity. CAPO is evaluated against a state-of-the-art genetic algorithm and mixed-integer programming solvers on six problems, including an acrobatic multirotor aircraft, a spinning space structure, cloth manipulation, and a soft robotic swimmer. On small-scale problems, CAPO finds comparable solutions with similar objective values in 2.5-218x less time than existing methods. On large-scale problems, CAPO is the only method capable of converging to a feasible solution, and it achieves actuator configurations that reduce the total number of actuators by 12%-27.5% compared to baselines.
High-Expansion-Ratio Deployable Structures for Long-Duration Space mission
Mitchell B. Fogelson*, Sawyer Thomas*, Giusy Falcone, Jeffrey I. Lipton, Zachary Manchester
(Big Sky, MT IEEE Aero 2023)
Artificial gravity is a crucial technology to enable long-duration human space flight. However, a kilometer-scale rotating space structure is needed to generate artificial gravity at rotation rates that can be tolerated comfortably by crew. Constructing such a structure with current technology would require many launches and significant in-space assembly. This work presents HERDS, High-Expansion-Ratio Deployable Structures, a hierarchical expansion mechanism that can deploy a kilometer-scale structure from a single launch. HERDS leverages a hierarchical combination of a Kresling mechanism and a pop-up extending truss (PET), a novel variant of the scissor mechanism. We show that HERDS designs achieve 4-11x better beam member aspect ratios than non-hierarchical Kresling or scissor mechanisms, resulting in a stiffer deployed structure. Furthermore, HERDS designs are shown in simulation to satisfy the necessary loading and structural constraints for supporting the Lunar Gateway mission with a factor of safety greater than 1.5 using existing launch vehicles. Our modeling and analysis is validated on a 1/10 scale prototype with a 50x expansion ratio.
EMI: An Expressive Mobile Interactive Robot
Yuhui You*, Mitchell B. Fogelson*, Kelvin Cheng, Bjorn Stenger
(Honolulu, HI CHI 2020)
In this paper, we explore how the emotional behavior of a robot affects interactions with humans. We introduce the EMI platform – an expressive, mobile and interactive robot – consisting of a circular diff-drive robot base equipped with a rear-projected expressive face, and omni-directional microphone for voice-interaction. We exhibited the EMI robot at a public event, in which attendees were given the option to interact with the robot and participate in a survey and observational study. The survey and observations focused on the effects of the robot's expressiveness in interactions with users of different ages and cultural backgrounds. From the survey responses, video observations and informal interviews we highlight key design decisions in EMI that resulted in positive user reactions.