Michi Jewett

Created a Deep-Learned, robotic fire suppression system to target fires early to minimize damage.
Led a team of 9 engineers for 22 weeks, coordinating meetings, design reviews, and Gantt charts.
Oversaw designing product for tolerance / demoing a robust prototype to robotics industry experts.
-------------
Flame-Imaging, Rapid Extinguisher (FIRE) is the project my team is working on for our Senior Capstone project in our Design of Robotics Systems course. Our team is made up of 9… Show more

Created a Deep-Learned, robotic fire suppression system to target fires early to minimize damage.
Led a team of 9 engineers for 22 weeks, coordinating meetings, design reviews, and Gantt charts.
Oversaw designing product for tolerance / demoing a robust prototype to robotics industry experts.
-------------
Flame-Imaging, Rapid Extinguisher (FIRE) is the project my team is working on for our Senior Capstone project in our Design of Robotics Systems course. Our team is made up of 9 engineers including myself, and I serve as the Project Leader, optimizing our workflow and delegating appropriate based on each person's strengths.

Following the formal top-down design process, we decided to create a solution that would modernize sprinkler systems in buildings to avoid document destruction and sprinkler replacement costs. Our fire suppression system will autonomously sense the location of the fire in the room and aim its nozzle at the fire such that it minimizes collateral damage and spreads resources where they are needed most.

As a team, we will be designing, CADing, and machining all of the parts necessary, as well as designing all the PCBs for the internal hardware. We will also be solving the nozzle locomotion and trajectory optimization problem in conjunction with computer vision and machine learning.

I am personally responsible for our dynamics model, sensor model, and discretization of the state and action space via collocation methods, as well as the optimization of these trajectories given some number of goal areas of fire that need to be suppressed. I am also assisting in the design of the computer vision component which involves thermal imaging and convolutional neural networks.

We will attend investor and shareholder meetings and present a finished product after 20 weeks of working on the project. Our project will be critiqued and reviewed by peers and professors, enabling us to create a solution robust enough to be a product for a startup after graduation. Show less

Solving trajectory optimization for locomotion of quadrupedal robots to automatically
determine gait sequencing, step timing, footholds, swing-phase motions, and six-dimensional
body motion over novel terrain, updated dynamically by a vision-based elevation map.

Using Deep Q-Learning to optimize Stochastic, Tube-Based Model Predictive Control and Simultaneous Localization and Mapping for the path planning of Autonomous Quadrupeds, Drones, and coordinated Multi-Robot systems in complex, unexplored environments.

Top-Down Puzzle Role-Playing Game written in Javascript. Playtime is 15-30 hours, depending on how clever you are! I have been working on this game since 2014 and it has gone through countless iterations and overhauls. Recently it has been released online.

Description:
"Travel up a tower known as Daedalus, comprised of ten uniquely-themed floors. There are an abundance of riddles, environmental challenges, calculation problems, pattern-based puzzles, and memory tests. Many puzzles… Show more

Top-Down Puzzle Role-Playing Game written in Javascript. Playtime is 15-30 hours, depending on how clever you are! I have been working on this game since 2014 and it has gone through countless iterations and overhauls. Recently it has been released online.

Description:
"Travel up a tower known as Daedalus, comprised of ten uniquely-themed floors. There are an abundance of riddles, environmental challenges, calculation problems, pattern-based puzzles, and memory tests. Many puzzles even involve legitimate engineering and computer science puzzles; do not worry though, the game explains everything any player would need to know. In fact, many players may end up learning something about circuit theory or assembly language without even knowing it!

You must ascend the tower: But, be sure to take heed and avoid flying too close to the sun.

What mysteries will you unravel in Daedalus?" Show less

Risk-Averse MPC via Visual-Inertial Input and Recurrent Networks for Online Collision Avoidance

This project proposed an online path planning architecture that extends the model predictive control (MPC) formulation to consider future location uncertainties for safer navigation through cluttered environments. Our algorithm combines an object detection pipeline with a recurrent neural network (RNN) which infers the covariance of state estimates through each step of our MPC’s finite time… Show more

Risk-Averse MPC via Visual-Inertial Input and Recurrent Networks for Online Collision Avoidance

This project proposed an online path planning architecture that extends the model predictive control (MPC) formulation to consider future location uncertainties for safer navigation through cluttered environments. Our algorithm combines an object detection pipeline with a recurrent neural network (RNN) which infers the covariance of state estimates through each step of our MPC’s finite time horizon. The RNN model is trained on a dataset that is comprised of robot and landmark poses generated from camera images and inertial measurement unit (IMU) readings via a state-of-the-art visual-inertial odometry framework. To detect and extract object locations for avoidance, we use a custom-trained convolutional neural network model in conjunction with a feature extractor to retrieve 3D centroid and radii boundaries of nearby obstacles. The robustness of our methods is validated on complex quadruped robot dynamics, which demonstrate autonomous behaviors that can plan fast and collision-free paths towards a goal point.

The project theory and execution was conceived by Alex Schperberg. Kenny Chen from the Laboratory for Embedded Machines and Ubiquitous Robots (LEMUR) was responsible for the CNN for object location and feature extraction. Stephanie Tsuei provided all support necessary from the UCLA Vision Lab and was responsible for the camera, which was integral to providing accurate visual stimulus. I was responsible for implementing the RNN which predicted the covariance, allowing us to not only take the shortest path, but the most certain one. Show less