Mike Stilman Ph.D.

Assistant Professor
Robotics and Intelligent Machines
School of Interactive Computing
Georgia Tech
801 Atlantic Drive
Atlanta, GA 30332


650.283.4284

Navigation Among Movable Obstacles [Project Webpage]

Abstract: We introduce a practical extension to navigation for humanoids and other dexterous mobile robots. The robot is permitted to reconfigure the environment by moving obstacles and clearing free space for a path. This work presents a resolution complete planner for a subclass of NAMO problems. Our planner takes advantage of the navigational structure through state-space decomposition and heuristic search. The planning complexity is reduced to the difficulty of the specific navigation task, rather than the dimensionality of the multiobject domain. We demonstrate real-time results for spaces that contain large numbers of movable obstacles. We also present a practical framework for single-agent search that can be used in algorithmic reasoning about this domain. [an error occurred while processing this directive]

M. Stilman, K. Nishiwaki, S. Kagami and J. Kuffner. Planning and Executing Navigation Among Movable Obstacles , Proceedings of the IEEE/RSJ International Conference on Intelligent Robots and Systems, Oct., 2006. [PDF][BibTex]

M. Stilman and J.J. Kuffner. Navigation Among Movable Obstacles: Real-Time Reasoning in Complex Environments, International Journal of Humanoid Robotics, Vol. 2, No. 4, December, 2005, pp. 479-504.[PDF][BibTex]



Humanoid Tele-Manipulation

Abstract: We present results of successful telemanipulation of large, heavy objects by a humanoid robot. Using a single joystick the operator controls walking and whole body manipulation along arbitrary paths for up to ten minutes of continuous execution. The robot grasps, walks, pushes, pulls, turns and re-grasps a 55kg range of loads on casters. Our telemanipulation framework changes reference frames online to let the operator steer the robot in free walking, its hands in grasping and the object during mobile manipulation. In the case of manipulation, our system computes a robot motion that satisfies the commanded object path as well as the kinematic and dynamic constraints of the robot. Furthermore, we achieve increased robot stability by learning dynamic friction models of manipulated objects.

M. Stilman, K. Nishiwaki and S. Kagami Humanoid Teleoperation for Whole Body Manipulation, Proceedings of the IEEE International Conference on Robotics and Automation (ICRA'08), May, 2008. [PDF][BibTex]

M. Stilman, K. Nishiwaki and S. Kagami Learning Object Models for Whole Body Manipulation, Proceedings of the 2007 IEEE Int. Conf. on Humanoid Robotics (Humanoids'07), November, 2007.[PDF][BibTex]




Compliant Motion Planning

Abstract: A study of randomized joint space path planning for articulated robots that are subject to task space constraints.This paper presents a representation of constrained motion for joint space planners and develops two simple and efficient methods for constrained sampling of joint configurations: Tangent Space Sampling (TS) and First-Order Retraction (FR). Constrained joint space planning is important for many real world problems involving redundant manipulators. On the one hand, tasks are designated in work space coordinates: rotating doors about fixed axes, sliding drawers along fixed trajectories or holding objects level during transport. On the other, joint space planning gives alternative paths that use redundant degrees of freedom to avoid obstacles or satisfy additional goals while performing a task. In simulation, we demonstrate that our methods are faster and significantly more invariant to problem/algorithm parameters than existing techniques.

Mike Stilman. Task Constrained Motion Planning in Robot Joint Space In Proceedings IEEE/RSJ Int. Conf. on Intelligent Robots and Systems IROS 07, Oct. 2007.[PDF] [BibTex]




Dynamic Biped Locomotion

Abstract: We explore the use of computational optimal control techniques for automated construction of policies in complex dynamic environments. Our implementation of dynamic programming is performed in a reduced dimensional subspace of a simulated four-DOF biped robot with point feet. We show that a computed solution to this problem can be generated and yield empirically stable walking that can handle various types of disturbances.

Mike Stilman, C.G. Atkeson, J. Kuffner, and G. Zeglin. Dynamic Programming in Reduced Dimensional Spaces: Dynamic Planning For Robust Biped Locomotion Proceedings of the IEEE International Conference on Robotics and Automation (ICRA'05), April, 2005. [PDF][BibTex]


Robot Dribbling

CS225A (2001): I worked with Anita Lillie and Drew Baglino to create a ball dribbling controller for a Puma 560. We replaced the end effector with a compliant half-cylindrical tube that structurally kept the ball in a plane. The feedback controller gathered information from a three axis force torque sensor and determined appropriate behavior for the arm. The controller consisted of two modules:
  • Hit the ball down on contact
  • Move sideways to give the ball spin
The latter control was particularly exciting since it used friction to move the the ball back towards the center. This made is possible to return to vertical dribbling from states that would be unrecoverable under the assumptions of rigid body dynamics.

Video: [Flash]




Robot Catch

(2002) We looked into the problem of two Puma arms performing a throw and catch task. The catching robot is not aware of the intentions of the throwing robot and must catch the ball via visual servoing. The relative proximity of the robots requires early detection and extrapolation of the ball path for the catcher to reach the ball.

Vision was performed using a Small Vision System (SVS) stereo camera that extracted stereo information of the scene at 30fps. We localized the ball using a blob finding algorithm in the range data. The robot also used its force sensor to detect whether the ball had been acquired successfully.

By instructing the pitcher to throw randomly in a designated workspace area we achieved over 10 consecutive catches. After each trial the catcher autonomously returned the ball to the pitcher.

Video: [Flash]




Real-Time Stereo Tracking [Project Webpage]

CS225B: (2002) Seungbum Koo, Anton Dizhur and I developed an active contour algorithm that tracked multiple subjects in real time. The tracker mainted contours around targets such as people moving in a three dimensional lab setting. Our algorithm operated on disparity data acquired from an SRI Small Vision System camera.

Active contour models, such as those in the Snake algorithm, use intensity gradients from images to fit contours to underlying features of objects. Each snake-like contour has internal and external energy which it tries to minimize. External energy is the effect of gradients pulling the contours to a close fit of features. Minimizing internal energy keeps the contours tight and smooth.

Video: [Flash]



Artificial Intelligence

(2001-2003) Before my research in spatial reasoning I looked into robots that think logically. My work with Professor John McCarthy and his Formal Reasoning Group studied effects that actions have on the world. One result summarized how effects could be classified according to previous knowledge and feasibility. This gave an intuitive solution to the frame problem for planning expressed in first-order logic.

M. Stilman. Causality Revisited: Reifying Effects Stanford University Dept. of Computer Science, Formal Reasoning Group Tech. Report, Aug. 2001 [PDF]

I also participated in Professor Eyal Amir's project on partitioning and reasoning. The goal was to increase the efficiency of reasoning about sets of axioms by localizing problems to subsets of axioms. My contributions were in software development and algorithm optimization for partitioning tools.



Dynamic Balancing [Anybots Webpage]

ANYBOTS (2003) Between Stanford and CMU I joined a unique company started by Dr. Trevor Blackwell in Mountain View, CA. Anybots' goal is to develop dynamically balancing humanoid robots that can operate in human enviornments. When I joined Anybots, Trevor had built the pneumatic biped Dexter. We worked together on designing control algorithms to make the robot balance. The challenge was to provide long term stability against disturbances with inherently compliant pneumatic actuators.

During my time at Anybots we achieved stable balancing and began initial steps towards walking. Recently, however, Trevor has succeeded in prolonged walking control of the biped. Please visit the Anybots webpage for these exciting new results.



3D Game Development

CS248: (2003) Mike Liu and I set out to create the ultimate 3D arcade game for the PC. This action packed project engages the player in motorcycle combat where riding, performing stunts and destroying enemies all contribute the score.

Creating a video game was educational and challenging. The complete project involved modeling the physics of motorcyle riding, weapon balistics and particle systems for explosions. Each component required attention to both physical realism and intuitive game play. Furthermore, we designed a control system for smooth, critically damped camera motion that would follow the rider and provide additional distance/panning during turns and stunts.