In addition to my primary research, I've worked on a number of projects, both in graduate school and while I was an undergraduate. A short overview of some selected research and design projects is given below -- click the thumbnails or scroll down to find out more.
We designed and built an automated dart-shooting mobile robot. The robot used a camera to look for and locate targets, then aimed a pneumatic turret atop the robot, loaded the turret with a dart, and fired at the target until it either disappeared or the robot ran out of darts. The Wrecking Bot used an AmigoBot as the base and was run via an Arduino for turret control and an on-board laptop running Visual C++ with appropriate libraries: OpenCV for machine vision and Aria for the AmigoBot. The turret assembly was constructed using primarily laser cut plywood and a number of 3D printed parts, such as those comprising the gimbal used to aim the solenoid valve-actuated dart gun.
As the final project for this class, we were required to lead a fleet of five mobile robots through a terrain using a MATLAB simulation environment provided by the course instructors. One robot, the leader, had information about the nearest waypoint, while the others could only sense obstacles and neighboring robots, within some specified radius. The aim of the project was to use multi-agent control to guide the robots through a variety of terrain (guided by sequential waypoints seen by the leader agent), maintain contact, avoid obstacles, then find and re-activate a number of lost and disabled robots from within a designated search area. Once all the robots had been found, they were brought to a platform located at the final waypoint and organized into a specific formation. The terrain forced the team to do several tasks, such as squeeze through a tunnel, avoid field obstacles, split and merge around obstacles, and drive the agents in a pre-defined formation. The figure on the right shows some screenshots of the robots in action. The agents are represented by circles-- red is a the leader, and blue circles are followers. The yellow circle is a waypoint, and green lines indicate edges between two agents; i.e., robots connected by green lines can sense each other's presence.
Screenshots from the simulation. Left: the robots are led through a field of obstacles. Right: The robots search a predefined space (shown in yellow) for lost agents (seen in black).
A model predictive controller (MPC) was used to couple human control inputs with an automatic controller for shared control of a quadruped rescue
robot, an application of Dr. Chipalkatty's dissertation work on MPC for shared control.
I worked with Dr. Chipalkatty to implement his control formulation on the Compact Rescue Robot testbed, as an instance of a semi-autonomous user interface. The formulation used the MPC to guide human controlled leg positions to satisfy state constraints that correspond to static stability for the robot. A hybrid control architecture was developed to implement a gait that guarantees stable foot-placements for the robot. The algorithm was applied to a simulation of a quadruped rescue robot with human input provided through haptic joysticks.
Left: a screenshot of the robot as the operator moves the leg. Right: Corresponding constraint enforced by MPC.
We designed a control system for a gaming chair with position and vibration haptic feedback in MATLAB and Simulink. The chair design used a hydraulic actuator for vertical position control, electric servomotors for pitch and roll, and a shaker for vibration feedback. Control was provided using a Linear Quadratic Regulator (LQR) coupled with a Kalman Filter for estimation in the presence of noise, resulting in a Linear Quadratic Gaussian (LQG) controller for the entire platform. Results were shown in simulation (in Simulink) and on hydraulic actuator assembly used to demonstrate hardware results for the vertical degree of freedom.
The hydraulic actuator (left) controlled using the LQG controller via a target computer running Simulink/xPC Target.
The development of a synthetic flexible jet would provide a novel method of propulsion for both soft-bodied and sub-aquatic vehicles. Similar to the jellyfish and squid in its structure and function, such a device requires substantial symmetric radial contraction of the main body. Using the cross section and mechanics of a Polyorchis Pencillatus hydromedusa as an inspirational tool, the basic jet is designed using a combination of cast silicone rubbers. The geometry and material ratios can subsequently be optimized through the development of appropriate physical and computational models. While the successful results of this structure are easily observed in nature, synthetic replication is difficult, as the material properties of the jellyfish muscle display minimal elastic moduli, up to 3 orders of magnitude lower than those feasibly achieved by elastomer molds. Accordingly, the geometry of the actuator must be altered to minimize the required actuating force.
The primary goal of this work was to construct a soft, cylindrical structure capable of halving its diameter given a reasonable actuating force. Moreover, the interior boundary of the structure needed to remain as cylindrical as possible to ensure smooth fluid flow for the jellyfish-inspired jet application. Through careful observation, modified casting methods, and alteration of the observed geometry, a jet design was developed that achieves a 50% reduction in the diameter of the inner ring. Additionally, two models were developed: simple radial spring model was used to balance the forces, and a finite element analysis provided a more robust simulation for a variety of design and actuating force variations. Results were demonstrated in simulation and on a prototype of the jet cross-section.
Comparison of elastomer molds (top) to jellyfish cross sections (bottom). The relaxed view is shown on the left, while the right view is the contracted version.
Our senior design project clients were doctors at Brigham and Women's Hospital and CIMIT who were concerned with applying external mechanical stimuli, as an additional mechanical factor, to encourage and even accelerate the process of wound healing. In a previous study conducted by our clients, the effects of uni-axial loading wound healing were tested on a rodent. The study showed that cyclic and continuous loading of the wound site has led to an increase in vasculature and in new blood vessels; i.e., that mechanical stimuli has a positive effect on the process of wound healing.
Based on the success of rodent model study, our clients wanted to pursue a mechanical design that is similar in function to the previous device and can be used in a clinical setting. The device needed to provide mechanical stimuli (tension/compression) to a skin surface and improve upon the single-axis research design by (1) separating the on-skin mechanism from the driving mechanism, (2) finding an efficient way to attach the on-skin mechanism to the skin, and (3) designing the device to treat head wounds by accommodating different contours.
The resulting design uses linear actuators to actuate an on-skin device via wires. The on-skin mechanism is composed of four attachment feet, four slotted rods, four attachment rods, and a centerpiece. In application, each attachment foot is attached to the skin surface using a medical adhesive. Each attachment foot slides on the slotted rod, and the slotted rod is connected to one end of the attachment rod. The other end of the attachment rod is anchored to the centerpiece using a set screw. Each attachment foot contains a load cell for force measurement, a compression spring for restoring force, and a cotter pin. For each of the attachment feet, one end of a thread is directly attached to the load cell, and the thread runs through the slotted rod and the attachment rod.
The other end of the thread is connected to the linear actuator assembly. By extending or retracting the actuator, the tension of the thread can be decreased or increased, respectively. By retracting the linear actuator, the tension of the thread will cause the load cell to deflect, and the load cell will force the attachment foot to slide along the slotted rod. The device is run via a LabVIEW program that interfaces with device through a data acquisition board.
For clinical application, our design would need to be modified to improve the portability of the on-skin mechanism and the reliability of the control system.
A view of the on-skin platform. The clear white elastomer, seen below one of the pads on the left, is used to emulate the behavior of skin.
Component view of the prototype with one foot connected to the corresponding actuator: (1) Device attached to elastomer pads in lieu of skin, (2) Off-skin/Actuation platform, and (3) NI DAQ board for interface with LabVIEW, running on laptop (not shown).
The long-term goal of this research was to simulate the propagation of pollution from automotive exhaust. To do so, atmospheric waves were used to create a two-dimensional model of the atmosphere that can solve the Shallow Water Equations (SWE), a set of partial differential equations used to simulate wave propagation when the horizontal dimensions are of much greater magnitude than the depth. This allows for the simulation of several different types of waves; most commonly atmospheric or oceanographic. A Weighted Essentially Non Oscillatory (WENO) scheme was used to solve the SWE. This approach eliminates oscillations resulting from discontinuities, and is more efficient than comparable solvers. Additionally, since the aim was to model pollution in the atmosphere, the SWE had to be solved on a globe. However, use of geographic coordinates can lead to challenges at poles of a sphere (as seen in the figure on the right); instead, the cubed sphere was suggested as an improved approach. The cubed sphere projects six square surfaces onto a cube, then used boundary conditions to project from one square to another.
The project was a dual effort with John Robinson, a student in a parallel REU at Technische Universitädt Darmstadt, and under the supervision of our faculty and graduate student advisers. We explored usage of Lax-Wendroff and Weighted Essentially Non-Oscillatory (WENO) schemes to solve the equations in Cartesian Coordinates. Further, proper methods and gridding techniques for solutions in spherical geometry, such as the cubed sphere, were examined for an arbitrary case.
Simulation of drop/ripple; comparison between Lax-Wendroff and WENO approaches after an equal number of iterations. The WENO result is smoother and free of oscillation.
Left: visualization of problems that result from solving SWE in geographic coordinates. Right: Cubed sphere concept.
As a first look into engineering design, Tufts sophomore mechanical engineers participated in a design contest using a manually triggered device. Devices faced off head-to-head in a circular arena. The arena was split into two semi-circles, and bounded by an outer and an inner ring. Each device started entirely between these rings, such that they bordered an initially empty circle in the center of the arena. A small ball was then placed in the center of the arena. The objective of the contest was to be the first to get the ball into the opponent's zone (i.e., across the inner ring on the opponent's half). Devices were triggered by a single manual motion, and were constrained in size and cost, and were allowed to attack the opposing device.
We employed an aggressive approach that used two rattrap powered rulers to attack the opponent's device. A spring-loaded dart fired at the ball initially, and two mousetrap powered arms swept the base as a second, more encompassing ball-moving method. A Schwarzenegger-emblazoned wall was used to limit the ball travel by the opponent.
This project was primarily a fun initial engineering challenge; the full-scale offensive approach was straightforward but effective, and the use of rat- and mousetraps and manual triggers was a neat introduction to reliable and user-friendly actuator design (it's not fun to get hit by rattraps). Plus, it was the closest academic excuse we had to build a battle-bot replica, so we rose to the occasion.
The device in the triggered position.
