Dr. Matei Ciocarlie




current position:
doctoral advisor:
email (still active):
Ph.D. Graduate
Sr. Research Scientist, Google Inc.
Prof. Peter Allen
Robotics Lab, Computer Science Department, Columbia University

Doctoral Thesis



My thesis introduces new methods for enabling the effective use of highly dexterous robotic hands, interfacing with the upcoming generation of neurally controlled hand prostheses, and designing a new class of simple yet effective grasping devices based on underactuation and mechanical adaptation. These methods share a common goal: reducing the complexity that has traditionally been associated, at both computational and mechanical levels, with robotic grasping in unstructured environments.

In this thesis, we propose using low-dimensional posture subspaces for dexterous or anthropomorphic hands. Human user studies have shown that most of the variance in hand posture for a wide range of grasping tasks is contained in relatively few dimensions. We extend these results to a range of robotic designs, and introduce the concept of eigengrasps as the bases of a low-dimensional, linear hand posture subspace. We then show that a grasp synthesis algorithm that optimizes hand posture in eigengrasp space is both computationally efficient and likely to yield stable grasps.

Algorithmic approaches to low-dimensional grasping can enable computationally effective algorithms and interaction models. Hardware implementations have the potential to reduce the mechanical complexity and construction costs of a hand design, using concepts such as underactuation and passive mechanical adaptation. Instead of complex run-time algorithms, hand models in this class use design-time analysis to improve performance over a spectrum of tasks. Along these directions, we present a set of analysis and optimization tools for the design of low-dimensional, underactuated hands. We focus on tendon-based mechanisms featuring adaptive joints and compliant fingertips, and show how a number of design parameters, such as tendon routes or joint stiffnesses, can be optimized to enable a wide range of stable grasps.

  • Dissertation. Low-Dimensional Robotic Grasping: Eigengrasp Subspaces and Optimized Underactuation, [pdf, 8.9 MB]
Selected Projects
Low dimensional hand control using Eigengrasps

One difficulty in understanding human hand control is the large number of degrees of freedom (DOFs) involved. This flexibility gives rise to an enormous set of possible hand configurations. The high dimensionality of the control space also explains the difficulty in creating effective control algorithms for all but the simplest artificial hand designs.

One possible explanation for human efficiency in selecting appropriate grasps assumes that humans unconsciously simplify the large search space through learning and experience. Recent advances in neuroscience research have shown that control of the human hand during grasping is indeed dominated by movement in a configuration space of highly reduced dimensionality. In my work, I extend this concept to robotic hands and show how a similar dimensionality reduction can be defined using a number of basis vectors called eigengrasps. This framework can be used to derive optimization algorithms that simplify the task of finding stable grasps even for highly complex hand designs. Furthermore, it offers a unified approach for controlling different hands, even if the kinematic structures of the models are significantly different.
  • Hand Posture Subspaces for Dexterous Robotic Grasping, M. Ciocarlie and P. Allen, Intl. Journal of Robotics Research 28(7), 2009 [link]
  • Dimensionality Reduction for Hand-Independent Dexterous Robotic Grasping, M. Ciocarlie, C. Goldfeder and P. Allen, IROS 2007 [pdf, 0.4 MB]
Image: planning results obtained by searching the eigengrasp space for stable grasps. The planning method uses a unified treatment for all robotic hand models in the image, even though the kinematic specifications are significantly different.
Optimization of underactuated hand designs

We present a method for analysis and optimization of tendon-based underactuated adaptive hands. We have integrated the co-actuation and compliance constraints, together with contact friction constraints, into a quasistatic equilibrium formulation. Using this model, we can build a solvable optimization problem to compute the hand design parameters that provide the best performance over a large set of grasping tasks. We believe that, for the class of adaptive underactuated hands, the on-line grasp planning effort, traditionally carried out under tight time constraints and requiring extensive sensing capabilities, can be replaced by off-line optimization increasing hand performance over many grasping scenarios. As a concrete example, we have analyzed a simplified single-tendon gripper model, and have used the results of the optimization to construct a prototype capable of a wide range of grasps.

  • A Constrained Optimization Framework for Compliant Underactuated Grasping, M. Ciocarlie and P. Allen, Workshop on Underactuated Grasping 2010, [pdf, 1.3MB]
  • Data-driven Optimization for Underactuated Robotic Hands, M. Ciocarlie and P. Allen, ICRA 2010, [pdf, 1.6MB]
  • Design and Analysis Tool for Underactuated Compliant Hands, M. Ciocarlie and P. Allen, IROS 2009, [pdf, 0.2MB]
Grasping and manipulation using soft fingertips



The ability to create stable, encompassing grasps with subsets of fingers is greatly increased by using soft fingertips that deform during contact and apply a larger space of frictional forces and moments than their rigid counterparts. This is true not only for human grasping, but also for robotic hands using fingerpads made of soft materials. Here are examples of dynamic simulations of an anthropomorphic hand using soft fingertips for performing manipulation tasks. In order to achieve computational rates needed for dynamic simulation, I am using an analytical soft finger model that uses general expressions for non-planar contacts of elastic bodies to account for the local geometry and structure of the bodies in contact.
  • Soft Finger Model with Adaptive Contact Geometry for Grasping and Manipulation Tasks, M. Ciocarlie, C. Lackner and P. Allen, World Haptics 2007 [Best Student Paper Award] [pdf, 1.23 MB]
Another approach to building soft finger contact models uses Finite Element Analysis (FEA) to explicitly compute fingertip deformation. This method is computationally expensive, but makes no assumptions regarding the geometry of the fingerpad and the results are very accurate from a physical point of view. It can also be used to compute the deformation of the fingertip in response to frictional forces and moments. We have applied this method for performing grasp analysis on pinch grasps using robotic hands equipped with fingers coated in a thin layer of rubber. More details, including the grasp analysis method, quality metrics as well as the finite element method used, can be found in:
  • Grasp Analysis Using Deformable Fingers, M. Ciocarlie, A. Miller and P. Allen, IROS 2005 [pdf, 0.6 MB]
Images: snapshot during the dynamic simulation of a pinch grasp using an analytical soft finger model (top); robotic hand equipped with with fingers coated in a thin layer of rubber (middle); fingertip deformation as seen from inside a transparent cube, computed using FEA (bottom)
GraspIt!

I am currently using and continuing to develop the GraspIt! simulator, originally built by Andrew Miller in the Robotics Lab at Columbia. This allows all the methods and tools presented above to be assembled into a complete computational model of a given grasping mechanism. Iterative refinements can be added to the models as we develop new insights into both robotic and human grasping. Development of the tools proposed in my research is aimed towards accurate modeling of new and/or more complex hand designs, including the human hand. Furthermore, GraspIt! can serve as a test bed for performing functional comparisons between such models.

GraspIt! is currently available for download. The publicly available release does not include support for soft fingertips or human hand models, however it has a large number of features extremely useful for grasp analysis and simulation. If you are already using this simulator and have any questions regarding it, or need help with particular aspects of the code, you can send me an email and I will do my best to provide support.
Physical-based dynamic simulation


I have developped a Finite Element Method based engine for dynamic simulation of deformable objects (top image shows this engine used to compute the deformation of a box-shaped object under the effect of gravity). This software can be used as a C++ library, and also as a stand-alone application as it includes an OpenGL-based  visualization component. The engine can currently use three simulation methods:
  • Newton iteration method converging to the equilibrium position of the system
  • Newmark step based method, used for dynamic simulations and computing the velocity and accelerations at the vertices of the deformable mesh
  • modified Newton iteration accounting for frictional contacts against planar surfaces. If the direction of relative motion at the contact is specified, this will compute the effects of contact normal forces as well as friction on the vertices of the deformable mesh.
This engine has been used for studying robotic fingertips (as described above) as well as human fingertips. The bottom image shows an anatomically correct fingertip model (with outer surface as well as the inner bone obtained from medical images) deforming under pressure from a planar surface.
Computational tools for modeling and visualizing historic and archaeological sites


Recent technological advances have greatly facilitated the process of obtaining digital data characterizing historical and archaeological sites. My interest is in using high resolution range scanners for capturing the three-dimensional structure of historical buildings. As a member of the Romanesque Churches of the Bourbonnais project, I have worked on acquiring complete three dimensional models of eight 12th century churches in the Bourbonnais region of France. These models, acquired using a Leica range scanner and a digital theodolyte, contain up to 15 mil. points for each building and can be used for visiting and studying the churches in virtual reality environments. Since both the interior and exterior of the building can be scanned and registeres into a single coordinate system, these models enable the creation of complete floor plans and cross sections, wich can also be used for studying the structural faults of the buildings.