Hiro's Research

Overview

"Find the shortest path for an unmanned vehicle to get to the goal while limiting the probability of crashing into obstacles to 0.01%"

"Minimize the supply of non-renewable energy for an electrical grid while limiting the probability of blackouts to 0.0001% per year"

I am developing algorithms that efficiently solve these problems. They are called joint chance-constrained finite-horizon optimal control problems: finding an optimal control sequence that minimizes a given cost function while limiting the probability of constraint violations to a user-specified risk bound. Many real world systems, such as unmanned vehicles and electrical grids, are exposed to significant uncertainty. For example, the velocity of unmanned aerial vehicles are uncertain due to turbulence; the demand for energy is uncertain due to unpredictable human behavior. My algorithms can explicitly take into account such stochastic uncertainties (typically modeled as Gaussian distribution) in environment and plant dynamics.

The key idea behind my algorithms is risk allocation. The cost can be minimized by optimally distributing the risk to each constraint. Intuitively, it corresponds to optimizing who, when, and where to take risk. My algorithm optimizes risk allocation and control sequence together by solving a convex optimization problem when the set of state constraitns are convex. The Non-convex Iterative Risk Allocation algorithm (NIRA) extends the risk allocation approach to problems with non-convex state constraints. By exploiting the problem structure, the NIRA algorithm achieved significant speed-up compared to naive Branch-and-Bound approach without making any compromise in solution optimality. I also developed the Market-based Risk  Allocation algorithm (MIRA) that solves chance-constrained optimal control problem for multi-agent systems in a decentralized manner. MIRA results in exactly the same optimal solution as the centralized optimization approach by decoupling the optimization problem through dual decomposition. Its solution process has an interesting analogy to an economic price adjustment process called tâtonnement.

Publications

Please see the publication list.

Applications

Personal Aerial Transportation System

(Image courtesy of Boeing Research and Technology)

Early work on my research has provided successful grounded demonstrations. For the past 3 years, the Boeing Company has funded my research to develop and use this algorithm to autonomously operate the personal transportation system (PTS). PTS consists of a fleet of personal aerial vehicle (PAVs) which serve as “flying taxis” to provide individuals with fast aerial point‐to‐point transportation. Since users of the PTS typically do not have a skill to pilot aircraft, the PAVs must be controlled autonomously and reliably. My algorithm, which can minimize the travel‐time while limiting the risk of failure, is the key to the efficient and safe operation of PTS. During this project I successfully demonstrated the capability of my algorithm to autonomously control aerial vehicles in a high‐fidelity simulator.

Sustainable Conencted Home

I am collaborating with MIT Mobile Experience Lab to design a smart residential building called Sustainable Connected Home. A fully realized, functioning prototype of Sustainable Connected Home will be built in Trento, Italy. The south façade of the Sustainable Connected Home is made of electrochromic glass, whose opacity can be controlled. My algorithms autonomously control the solar heat input to the house by modifying the opacity of windows in order to minimize the usage of heating systems while at the same time limiting the risk of overheating or overcooling the room, despite uncertain weather.

Codes

+ MIRA (Market-based Iterative Risk Allocation) - Optimizes the control sequence and risk allocation of a multi-agent system under uncertainty.