Research Topics

I have been working on a variety of fields, ranging from robotic exploration of asteroids and Mars, to formation flying, future urban mobility systems, distributed coordination of multi-robot networks, intelligent prosthetic interfaces, and bio-inspired robotics. Please find below more details (updated June 2011).

All Access Surface Mobility on Small Bodies

Hopping/Tumbling robot 

The last Planetary Science Decadal Survey (published on March 7, 2011) has prioritized missions to small bodies (e.g., asteroids, comets, Phobos, Deimos, and other Kuiper belt objects). Their in-situ exploration, in particular, would be pivotal to shed light on the origin of the Solar System, and would be instrumental in developing technologies for the future manned exploration of Mars. As a consequence, there is currently a strong interest in developing robotic platforms capable of fast and precise mobility on the surface of small bodies.

The aim of this line of research is to develop a unique spacecraft/rover hybrid capable of precise all surface mobility on small bodies, and to study mission scenarios enabled by this concept. This project is in collaboration with J. Castillo (JPL), I. Nesnas (JPL), J. Hoffman (MIT), and R. Binzel (MIT).

Future Urban Transportation Systems

Mobility-on-Demand system 

In 2001, personal urban mobility in the US resulted in more than 3.5 trillion urban miles traveled by private cars, representing 75% of total car travel in the US. This figure, coupled with the fact that by 2030 the total population living in urban areas will jump from the current 40% to more than 60%, implies that the demand for personal urban mobility will increase to formidable levels. The demand for roads and parking space will dramatically increase, while the available urban land will continue to decrease. The result is that private automobiles are an unsustainable solution for the future of personal mobility in dense urban environments.

One of the leading emerging paradigm for future urban mobility is represented by one-way vehicle sharing in the form of Mobility-on-Demand (MOD) systems, which provide stacks and racks of light electric vehicles at closely spaced intervals throughout a city: when a person wants to go somewhere, she simply walks to the nearest rack, swipes a card to pick up a vehicle, drives it (or is driven) to the rack nearest to her destination, and drops it off (image courtesy of MIT Smart Cities Group).

In collaboration with E. Frazzoli (MIT), S. Smith (U. of Waterloo), K. Treleaven (MIT), and D. Rus (MIT), I have investigated MOD systems along three dimensions:

  • Devised a polynomial-time, asymptotically optimal algorithm to solve the Stacker Crane Problem (which consists in finding minimum-length tour through pickup and delivery pairs in the Euclidean plane). [Paper]

  • Devised a polynomial-time, constant-factor algorithm for the solution of the dynamic version of the Stacker Crane Problem (where pickup and delivery pairs arrive according to a Poisson process). [Paper]

  • Developed a real-time policy to rebalance robotic cars in MOD systems [Paper].

Dynamic Vehicle Routing for Robotic Networks

DVRP with priorities 

The last decade has seen an increasing number of application domains where networks of uninhabited vehicles (UVs) are required to fulfill tasks that arise dynamically in time, are spatially distributed over an environment, and possibly require some type of additional on-site service. UAV systems, robotic environmental monitoring, and sensor networks are clear examples. The problem is to devise real-time routing algorithms to enable the UVs to search, identify, allocate, prioritize, and plan paths. (The figure on the left illustrates the problem: Panel #1: vehicles are assigned to tasks and select routes. Panel #2: the problem is how to re-allocate and re-plan routes when new tasks appear.)

In collaboration with E. Frazzoli (MIT), F. Bullo (UCSB), and S. Smith (U. Waterloo), I developed a novel approach for the solution to this problem, called algorithmic queueing theory, which relies upon methods from queueing theory, combinatorial optimization, and stochastic geometry (Paper). Algorithmic queueing theory consists of three basics steps: 1) queueing model of the problem and analysis of its structure; 2) establishment of fundamental limitations on performance, independent of algorithms; and 3) design of algorithms that are either optimal or constant-factor away from optimal. By using algorithmic queueing theory, I

  • designed adaptive and distributed control laws for routing UVs in dynamic and uncertain environments, [Paper],

  • devised polynomial-time, constant-factor routing algorithms for the case where tasks have deterministic deadlines on their waiting times, [Paper],

  • devised polynomial-time, constant-factor routing algorithms for the case with multiple priority classes of service demands, [Paper].

Distributed Control of Spacecraft Formations

Formation flying 

Significant interest in formation flying started to develop in the late 1990s, and today formation flying is a critical technology for many planned and future missions of NASA, the DoD, and ESA. Spacecraft formation algorithms can be divided into three main architectures: (i) Multiple-Input Multiple-Output (MIMO), in which the formation is treated as a single multiple-input, multiple-output plant, (ii) Leader-Follower, in which individual spacecraft controllers are connected hierarchically, and (iii) Cyclic, in which individual spacecraft controllers are connected non-hierarchically. Cyclic algorithms can distribute control effort more evenly, are generally more robust than MIMO algorithms, and can also be completely distributed. The two primary drawbacks of Cyclic algorithms are that the stability of these algorithms and their information requirements are poorly understood (a comprehensive and up-to-date survey on spacecraft formation flying can be found here).

Motivated by the discussion above, we have developed a class of Cyclic algorithms for formation flying, for which (i) a rigorous stability analysis is possible, and (ii) the information requirements are minimal. In particular, we studied control policies that only rely on relative measurements, since in deep-space missions global measurements may not be available. Our control algorithms have been tested on the International Space Station in 2008/2009. Specifically, we tested balanced circular formation for three Spheres satellites (videos of the tests in space can be found here). This is joint work with E. Frazzoli (MIT), J. Ramirez (MIT), and D. Miller (MIT). [Paper] and [Paper]

Sharing the Load in Mobile Robotic Networks

Equitable partition 

The design of cooperative control policies for the robots has to typically address three key challenges: (i) task allocation among the robots, (ii) service scheduling for each robot, and (iii) design of loitering strategies, i.e., strategies to adopt for robots with no assigned tasks. In general, these challenges are coupled. Therefore, devising an optimal, or at least provably efficient policy is often a difficult problem.

A natural way to reduce the complexity is to partition the workspace among the robots and then let each robot follow a certain set of rules in its own region. To what extent does this decoupling strategy affect optimality? In collaboration with A. Arsie (U. of Toledo), E. Frazzoli (MIT), and F. Bullo (UCSB), I have:

  • characterized specific scenarios where one can retain optimality, or at least some degree of optimality, under this partitioning scheme, [Paper],

  • designed partitioning algorithms that do not require any centralized computation (an important property for robotic networks comprising several robots that operate in an unknown dynamic environment). The design and analysis of such algorithms require a variety of techniques from control theory, computational geometry and algebraic topology. Their application can be found here. [Paper]

Bio-inspired Robotics for Planetary Exploration

Cockroach robot 

Planetary exploration requires autonomous robots able to perform missions in harsh and hazardous environments. Indeed, biology provides a wealth of inspiration: insects are able to transverse harsh terrains, to climb over obstacles, or even to walk upside down. Hence, my objective was to design a legged robot that imitates the mechanics and the behaviors of the insects, in particular of the cockroaches.

In order to replicate at least in part the extraordinary agility of cockroaches, I designed a six-legged robot, where each of the three pairs of legs has a unique design: front legs and middle legs have 3 degree of freedoms and a pantograph mechanism that facilitates the task of climbing obstacles, while the rear legs have 2 degrees of freedom and a piston-like design that guarantees powerful forward thrusting. The control system architecture has a two-level hierarchical organization: the low-level control layer is based on a CNN-based Central Pattern Generator (CPG); the CPG provides the basic rhythmic signals needed for locomotion. The high-level control layer is responsible for handling complex tasks like obstacle climbing and target pursuing, and is behavior-based. The key idea that we introduced is the formalization of a behavior as a Motor Map driven by an adaptive reward function. Experiments have shown that the robot is able to walk at the travel speed of 0.1 body length per second, and to successfully negotiate obstacles whose height is larger than the 120% of the height of the front part of the robot.

In collaboration with S. De Fiore and S. Sorbello, I also designed and built the control system of a mobile robot, named M6, whose purpose is the exploration of volcanos. M6 has six wheels which are coupled to the chassis by means of revolute joints. Each wheel is independently actuated. The robot uses a peristaltic motion to negotiate steep slopes.