Igor Mordatch
UC Berkeley Robotics

University of California, Berkeley
750 Sutardja Dai Hall
Berkeley, CA 98195
[last name]@berkeley.edu
+1 510 459 3152

Curriculum Vitae

About Me
I am a post-doctoral fellow working with Professor Pieter Abbeel at University of California, Berkeley. I received my PhD at University of Washington under the supervision of Professors Emanuel Todorov and Zoran Popovic. I previously completed a master's and undergraduate degree in Computer Science and Mathematics at University of Toronto under supervision of Professor Aaron Hertzmann. My research interests lie in the use of optimal control and machine learning techniques for robotics, computer graphics, and biomechanics.

Policy Learning with Neural Networks
An approach to control policy learning that simplifies the problem to an alternating sequence of trajectory optimization and neural network regression problems.

Interactive Control of Diverse Complex Characters with Neural Networks
Neural Information Processing Systems (NIPS Oral), 2015
Igor Mordatch, Kendall Lowrey, Galen Andrew, Zoran Popović, Emanuel Todorov
webpage   paper   video

Combining the benefits of function approximation and trajectory optimization
Robotics: Science and Systems (RSS), 2014
Igor Mordatch, Emanuel Todorov
webpage   paper   video

Contact-Invariant Optimization
Character control work focusing on automatically synthesizing behaviors that have complex contact structure, such as getting up, climbing, picking up objects, hand manipulation and so on.

Discovery of Complex Behaviors through Contact-Invariant Optimization
ACM Transactions on Graphics (SIGGRAPH), 2012
Igor Mordatch, Emanuel Todorov, Zoran Popović
webpage   paper   video

Contact-Invariant Optimization for Hand Manipulation
Eurographics/ACM Symposium on Computer Animation (SCA), 2012
Igor Mordatch, Zoran Popović, Emanuel Todorov
webpage   paper   video

Control of Physical Robots
Extension of trajectory optimization work applicable to control of physical robots.

Combining Model-Based Policy Search with Online Model Learning for Control of Physical Humanoids
Preprint, 2016
Igor Mordatch, Nikhil Mishra, Clemens Eppner, Pieter Abbeel
webpage   paper   video

Ensemble CIO: Full-Body Dynamic Motion Planning That Transfers to Physical Humanoids
International Conference on Intelligent Robots and Systems (IROS), 2015
Igor Mordatch, Kendall Lowrey, Emanuel Todorov
webpage   paper   video

Physics-based Trajectory Optimization for Grasping in Cluttered Environments
International Conference on Robotics and Automation (ICRA), 2015
Nikita Kitaev, Igor Mordatch, Sachin Patil, Pieter Abbeel
webpage   paper   video

Musculontendon Actuation
An application of Contact-Invariant Optimization to models with musculotendon unit actuators to animate human activities that are driven by the lower body.

Animating Human Lower Limbs Using Contact-Invariant Optimization
ACM Transactions on Graphics (SIGGRAPH), 2013
Igor Mordatch, Jack M. Wang, Emanuel Todorov, Vladlen Koltun
webpage   paper   video

Physics-Based Locomotion Control
A graduate project in creating simulated characters that can robustly perform a variety of locomotion tasks and generalize across a wide range of character types.

Robust Physics-Based Locomotion Using Low-Dimensional Planning
ACM Transactions on Graphics (SIGGRAPH), 2010
Igor Mordatch, Martin de Lasa, Aaron Hertzmann
webpage   paper   video

Feature-Based Locomotion Controllers
ACM Transactions on Graphics (SIGGRAPH), 2010
Martin de Lasa, Igor Mordatch, Aaron Hertzmann
webpage   paper   video

Stylized Animation
Extending the method of Image Analogies, this method creates temporally coherent stylized animations by example. To make the method art directable, it allows artists to paint keyframes that are used as constraints. The in-betweens calculated by this method maintain stylistic continuity and yet change no more than necessary over time.

Stylizing Animation By Example
ACM Transactions on Graphics (SIGGRAPH), 2013
Pierre Bénard, Forrester Cole, Michael Kass, Igor Mordatch, James Hegarty, Martin Sebastian Senn, Kurt Fleischer, Davide Pesare, Katherine Breeden
webpage   paper   video

Simulation-Based Watercolor
An undergraduate project in trying to replicate the look and motion of watercolor paint. The core is based on parts of MoXi, Computer-Generated Watercolor papers and uses MacCormack method for more accurate pigment advection. Paintbrushes were simulated at the level of individual hairs with a simplified torsion model. The applications are lacking at the moment, consisting of a basic interactive painting system and a simple automated image painting.

3D Navigation
At Autodesk Research, besides working on Nucleus simulation framework, I was involved in a project to improve 3D navigation experience for new users. The work has been incorporated into a surprisingly large number of Autodesk products and resulted in several publications:

Multiscale 3D Navigation
Symposium on Interactive 3D Graphics (I3D), 2009
James McCrae, Igor Mordatch, Michael Glueck, Azam Khan
webpage   paper   video

ViewCube: A 3D Orientation Indicator and Controller
Symposium on Interactive 3D Graphics (I3D), 2008
Azam Khan, Igor Mordatch, George Fitzmaurice, Justin Matejka, Gordon Kurtenbach
webpage   paper   video

Safe 3D Navigation
Symposium on Interactive 3D Graphics (I3D), 2008
George Fitzmaurice, Justin Matejka, Igor Mordatch, Azam Khan, Gordon Kurtenbach
webpage   paper   video

Real-Time Graphics
An entry for a one-person real-time graphics demo competition submitted while working at Autodesk. Done in three weeks' worth of evenings, it had to adhere to noir/horror theme. Environment and low resolution character were modeled and rigged in 3dsmax and detailed in mudbox. Small and basic shader-based engine was written. Using deferred shading design turned out to be a great decision and made for a very clean shadow and postprocessing implementation.

Miscellaneous Graphics
I suppose no list of graphics projects would be complete without a raytracer that involves reflective spheres and checkerboards...

A semi-automatic painterly rendering system that lays down strokes at a location and scale of the user's choosing. Can produce fairly interesting-looking images for about five minutes' worth of work.

Character Posing
A full-body FK/IK character posing system that treats the task as a nonlinear optimization problem. This is much cleaner than trying to adapt typical chain IK algorithms to multiple objectives and tree structures. Gradients are available, so the approach is also reasonably fast. Rather than set by hand, joint limits are recovered from training motions: for each joint, limits are a bounding volume in the tangent space of quaternions that encloses all observed orientations.

Spatial Keyframing
An undergraduate course project trying to extend the idea of spatial keyframing. The system was implemented as a 3dsmax plugin/script hybrid and provided great practical experience with scattered data interpolation, dimensionality reduction and global optimization. While it is questionable how well the system is suited for full-body animation, it could be very appropriate for niche tasks such as facial and hand animation.

Spatial Pose Trees: Creating and Editing Motions Using a Hierarchy of Low-Dimensional Control Spaces
University of Toronto Technical Report, 2007
Igor Mordatch, Patrick Coleman, Karan Singh, and Ravin Balakrishnan
paper   video

Interface Techniques for 3D Control of Spatial Keyframing
SIGGRAPH Posters, 2007
Igor Mordatch, Patrick Coleman, Karan Singh, and Ravin Balakrishnan

Painting and Animation
See my drawing blog for most recent artwork.