Projects (out of date)
A method to determine the optimal features for control of a powered lower-limb prostheses(EMBC May 2011)
Abstract—Lower-limb prostheses are rapidly advancing with greater computing power and sensing modalities. This paper is an attempt to begin exploring the trade-off between extrinsic and intrinsic control modalities. In this case, between electromyographic (extrinsic) and several internal sensors that can be used for intrinsic control. We propose a method that will identify the particular features, taken from two trans-femoral amputee and one trans-tibial amputee, during locomotion on varying terrain, that perfectly discriminate between locomotion modes. From this we are able to identify the source of the discriminability from a large-set of features that does not depend on the type of amputation. Also, we comment on the use of this algorithm in selecting the most discriminatory and least encumbering sensor/feature combination for transitions when the ground underneath the foot is unknown for trans-tibial amputees.
The Development of a Context and User Aware Robotic Ankle (December 2009 - Current)
The goal of this project is to develop algorithms that will make a robotic ankle context-aware. Users of robotic ankle users will be able to add and adjust existing control algorithms to fit their daily needs. Currently, to reduce energy expenditure in level ground walking a robotic ankle prostheses can provide power during late stance to assist in level ground walking for amputees. However, the most advanced robotic ankles available are not designed to handle more than walking on level ground and stair descent. To make the ankle context-aware I will have to create a new set of algorithms to tell the ankle the best state for a given context. This will be done by taking user commands through electromyographic sensors placed on a residual limb to switch between different states. Once in a particular state, the ankle will execute motor commands appropriate to that context; control of this sort is called extrinsic control. Electromyography (EMG) is a notoriously noisy signal and this limits the ability to use it for fine motor control. Our approach is to use the time when intrinsic control is functioning to maintain the EMG specific algorithms so that time-varying degradation in EMG specific algorithms can be controlled so they can be used in the future. These changes would improve the safety of the device, expand the vocabulary of motions to the user, and provide a more natural experience during usage.
Angular Momentum in Turning and Rapid Starts (September 2007 - May 2009)
Masters Research: Transients occur in human walking during a transition to, from, and between steady state walking and acts as an impulse destabilizing a gait cycle. Turns and accelerated starts are all common transients encountered and managed intelligently by humans everyday. As the population of elderly increases understanding balance control in healthy subjects becomes more important. Technologically, humanoid bipeds are rapidly becoming a more common part of our everyday life. Therefore, they must also be able to navigate our environments adroitly if they are to assist us in our daily living. This thesis takes biomechanical principals of angular momentum and applies them to healthy subjects in an e ort to elucidate human balance control strategies. Each transient task is unique and despite large segmental contributions to whole-body angular momentum during movement the whole-body angular momentum remains tightly regulated. A analysis of segmental contributions to the principal components explaining more than 90% makes clear the balance control strategy used by healthy humans.
A System for Recognition of Affect in Autism (Spring 2008)
People with Autism Spectrum Disorders (ASD) face a number of different challenges. These include social, emotional and physical challenges. Each challenge expresses itself differently in each person on the spectrum and the exhibited severity varies greatly from each person. The particular interest of this work is to understand some of the diversity of motor disorders occurring with people on the spectrum and develop technologies to promote more independent living by people on the Autistic Spectrum.
Control of the Compass Gait Biped via Weight Perturbation (Spring 2008)
The Compass Gait is a simplified model of biped walking. It has been shown there exist stable limit cycles for the passive dynamic walker. To increase the size of the basin of attraction it is possible to provide control at the hip and ankles, and design other energy shaping control to adjust speed and stability of the walker. In this paper I implement two controllers discovered by Goswami, et. al. that enlarge the basin of attraction to control a biped. Further, I go on to implement a Weight Perturbation algorithm to maximize the size of the basin of attraction for a small part of state space by optimizing the gains on the hip controller provided by Goswami.
Mechatonics and Prompt-Assisted Typing (Spring 2008)
Serveral different metholodologies have been developed to detect the onset of physical movement disorders in the past. Most research has thus far focued on detecting the onset of epileptic seizures. However, these ideas can be fruitfully applied to movement disorders in autism. This project aims to develop a wearable-device that detects the onset of muscular disfunction during the targetted tast of typing, and attempts to alieviate the discordination by providing vibractile feedback to the skin to relax muscles at an important point.
Stock Price Prediction using Gaussian Process Regression (2007)
Publications
M. Farrell, H. Herr, Angular Momentum Primitives for Human Turning: Control Implications for Biped Robots, IEEE - Humanoids Conference, December 2008
E.Martinez-Villalpando, H. Herr, M. Farrell
"Estimation of Ground Reaction Force and Zero Moment Point on a Powered Ankle-Foot Prosthesis", IEEE-EMBS Annual International Conference, 2007
M. Farrell, I. Masaki, B. Horn, "Vision Based System for Occupancy and Posture Analysis", MTL Annual Research Report, 2006
M. Farrell, "Theory of Theta-Schottky Groups and Their Fractal Dimensions", Thesis, Bard College at Simon's Rock, 2004
M. Farrell, "Solution to the Linked Circles Puzzle using Hopf Fibration", Bard College at Simon's Rock, 2004 (unpublished)
Coursework
MIT Graduate Courses:
Spring 2009: Statistical Learning Theory and Applications (9.520)
Fall 2008: Time Series Analysis and System Identification (2.687)
Spring 2008: Affective Computing(MAS.630), Underactuated Robotics(6.832)
Fall 2007: Optimization Methods(6.255), Machine Learning(6.867)
Spring 2006: Advanced Computational Imaging(6.881)
Fall 2005: Machine Vision(6.866)
Spring 2005: Advanced Partial Differential Equations (18.306)
Bard College at Simon's Rock: Real Analysis, Complex Analysis, Abstract Algebra, Fractal Geometry, Morse Theory, Discrete Geometric Group Theory, Topology, Knot Theory, Analytic Number Theory, Differential Geometry, Quantum Mechanics, Classical Mechanics