$$ % Define your custom commands here \newcommand{\bmat}[1]{\begin{bmatrix}#1\end{bmatrix}} \newcommand{\E}{\mathbb{E}} \newcommand{\P}{\mathbb{P}} \newcommand{\S}{\mathbb{S}} \newcommand{\R}{\mathbb{R}} \newcommand{\S}{\mathbb{S}} \newcommand{\norm}[2]{\|{#1}\|_{{}_{#2}}} \newcommand{\pd}[2]{\frac{\partial #1}{\partial #2}} \newcommand{\pdd}[2]{\frac{\partial^2 #1}{\partial #2^2}} \newcommand{\vectornorm}[1]{\left|\left|#1\right|\right|} \newcommand{\abs}[1]{\left|{#1}\right|} \newcommand{\mbf}[1]{\mathbf{#1}} \newcommand{\mc}[1]{\mathcal{#1}} \newcommand{\bm}[1]{\boldsymbol{#1}} \newcommand{\nicefrac}[2]{{}^{#1}\!/_{\!#2}} \newcommand{\argmin}{\operatorname*{arg\,min}} \newcommand{\argmax}{\operatorname*{arg\,max}} \newcommand{\dd}{\operatorname{d}\!} $$

Robotics and Control

Research in energy-based control, dynamic manipulation, locomotion, and coordinated robotic systems.

How can robots use their natural dynamics to move and manipulate objects?

Robots must work within physical constraints: limited actuation, changing contact, and the motion of the objects around them. Our research uses nonlinear control, geometric mechanics, and optimization to study these systems and design feedback for them.

Energy shaping and underactuated systems

An underactuated robot has fewer independent control inputs than degrees of freedom. This makes the relationship between its natural motion and its control inputs especially important.

Our work studies passivity-based control, which approaches feedback design through the system’s energy and its exchange with the environment. We also investigate data-driven methods for solving the equations that arise in energy-shaping controller design.

Representative work

Data-Driven Passivity-Based Control of Underactuated Mechanical Systems via Interconnection and Damping Assignment uses neural approximations to formulate the controller-design equations as an optimization problem.

Publication and Scholar search · Related learning research

Dynamic manipulation

Manipulation includes tasks in which an object is not continuously held in a rigid grasp. Tossing, catching, and batting require reasoning about the motion of both the robot and the object.

Our work includes a coordinate-free treatment of robotic pizza tossing and catching, as well as trajectory planning for robotic batting. These studies connect geometric descriptions of motion with task-specific control problems.

Nonprehensile manipulation

Tossing and batting release the object entirely. A second family of tasks never releases it and never grasps it either: pushing, toppling, and pivoting an object by making and breaking contacts in the right order. Here the contact sequence is the plan, and the planner has to invent it — including contacts that do not exist yet when planning begins.

Our current work asks what the discrete object should be in that setting. The answer we are pursuing keeps every candidate contact in the state as a bit vector, so that acquiring a contact becomes an ordinary planning action, and gates each release by a static support test.

Ongoing work

Nonprehensile manipulation → — the research thread, its status, and what is being measured.

Planning Through Contact → — an illustrated, figure-first walkthrough of the full construction.

Locomotion and coordinated motion

We study motion across systems ranging from walking models to groups of mobile robots. Representative topics include energy-efficient stabilization of a rimless wheel with a torso, optimal control of a biped, and formation control with connectivity and collision constraints.

Demonstrations

Explore the research through these demonstrations from the lab’s video collection.

Robust machine learning → · All publications →