mihailp @ seas.upenn.edu   +1-412-805-0210            Curriculum Vitae


Manipulation Planning


Computing efficient motions for manipulator arms poses a number of difficulties due to the complexity of the systems (often of high dimensionality) and challenging tasks (e.g. dexterous handling of objects amidst obstacles in uncertain environments). Ongoing work in this area includes leveraging continuous optimization techniques for motion planning with high dimensionality, sampling based planning, computing effective grasping strategies, as well as developing sets of motion primitives for arm motion and grasping.

Acknowledgements: DARPA Autonomous Robotic Manipulation (ARM-S)
DARPA logo   

Covariant Optimization

covariant optimization
Covariant optimization methods have been applied to generating smooth, collision-free trajectories, even amid dense obstacles. In this work, we achieved significant reduction in planning runtime by performing a variety of pre-computation off-line. We continue to seek other ways of improving planning performance in complex environments with clutter.


Grasp Planning

Grad fidelity
We seek to develop grasping strategies for an array of complex objects, including the grasps that allow simultaneous transport and manipulation of the object. Special interest is placed on learning grasping strategies that generalize well to objects that were not seen previously. Performance improvement is achieved by pre-computing grasp candidates for certain known objects. We also study pre-manipulation strategies and other ways of interacting with the objects before grasping to improve subsequent grasp quality.

Learning from Experience

learning from experience
By analyzing the sets of motions that are repeatedly computed and successfully executed on the robot for manipulating physical objects, we attempt to enable the robot to learn from its experience in the environment. This on-going effort appears promising in further reducing planning runtime by partially reusing previous computation, while also improving the quality of motions.