…this is horribly outdated…
Full Body Control Strategies for Dynamically Stable Humanoid Robots
With Kasemsit Teeyapan and Mike Stilman
June, 2009 – September, 2009
Wheeled humanoids and mobile manipulators have become a compelling alternative to bipedal robots due to the potential for increased stability and safety in human environments. Dynamic stability requires greater complexity in control, yet it also leads to greater robot capabilities. In this work, we investigate how whole body articulation can yield improved performance in navigation and naturally leads to human-like emergent behaviors.
In this work, we focus on Golem Krang, a novel two-wheeled humanoid robot constructed in our lab. We model the system as a planar mechanism consisting of two massive links connected by a revolute joint, with the lower link connects to a wheel with sufficient ground friction to maintain no-slip contact. Using this model, we demonstrate that upper body articulation greatly improves dynamic balancing as well as three simple robot tasks of standing, acceleration and deceleration.
Robot Limbo: Optimized Planning & Control for Dynamically Stable Robots Under Vertical Obstacles
With Kasemsit Teeyapan and Mike Stilman
June, 2009 – September, 2009
Search and rescue robots that enter disaster areas will need to go around fallen debris, go over rubble and go under partially collapsed supports and hanging wires. The former two types of navigation can be solved by existing algorithms in motion planning with stable and adaptive control. However, passing under obstacles remains a challenging open problem. This work explores the ability of a two-wheeled robot in generating motions to move or duck under obstacles. Generally, obstacles are immovable so statically stable robots might not easily accomplish this particular task. With the dynamically stable capability, the robot’s choice is either to lean forward or backward while it is beneath the obstacle. In the later case, the action is similar to the so- called limbo, a West Indian dance in which the dancer leans backward to go under the fixed-height stick.
Given the parameters of the mission such as the dimensions of the obstacle, our method utilities a LQ/IO-linearization hybrid controller in a sequential composition and uses stochastic optimization to automatically compute an optimized selection of controller gains as well as switching time for the overall torque. We demonstrate this system through numerical simulations, validation in a physics-based simulation environment, as well as on a novel two-wheeled platform. The results show that the generated control strategies are successful in mission planning for this challenging problem domain and offer significant advantages over hand-tuned alternatives.
Golem Krang: A Dynamically Stable Humanoid Robot for Mobile Manipulation
With Mike Stilman and many others
January, 2009 – Present
Golem Krang is a new balancing humanoid robot under construction at the Humanoid Robotics Lab at Georgia Tech. This effort is led by Mike Stilman, with numerous graduate and undergraduate students as a part of the Spring 2009 course “Building Humanoid Robots”. The entire system was conceptualized, planned, designed, and constructed in six months, with various on-going research efforts in planning, control, perception, etc.
The series name Golem is derived from Jewish folklore golem, a creature created by Judah Loew ben Bezalel, the late 16th century chief rabbi of Prague known as the Maharal, to defend Prague ghetto from anti-Semitic attacks. The model Krang is named after the super-villain who appeared in the Teenage Mutant Ninja Turtles universe, who builds robots.
Golem Krang is designed to have a Segway-style balancing base with a pair of 24V half horsepower motors mated to a 15:1 gearbox. Sensing is provided by a 6-DOF inertial measurement unit and a pair of stereo cameras. Two Schunk LWA3 7-DOF arms, each with Schunk SDH 3-finger hand, are used for manipulation. The software platform sits on a Pentium-M based computer running a real-time Linux kernel and power is supplied by eight 3.2v LiFePo batteries connected in parallel. Golem Krang weights approximately 250lb.
Robot Jenga: Autonomous and Strategic Block Extraction
With Philip D. Rogers, Lonnie T. Parker, Douglas A. Brooks, and Mike Stilman
Fall, 2008
Jenga is a popular game, marketed by Hasbro, consisting of 54 wooden blocks constructed into a tower of three blocks per level. Players take turns identifying and removing loose blocks, and placing them on top of the tower. Tower instability, uncertainty in pressure distribution, and imprecise force control are all known to be difficulties for human players. A robot manipulator that plays Jenga faces an even greater challenge: the robot must also recognize and localize Jenga pieces as well as track their unexpected motion.
The main contribution of the paper is the development and evaluation of an autonomous and strategic Jenga system, with particular considerations to the planning of the block extraction process. In this work, we use a three-stage planner for block extraction which explicitly tracks the extraction history and simulates tower stability using a physics engine. Using an active contours-based block pose estimation and optical flow-based tower movement detection, we do not assume a structured environment such as painted blocks used in the Kroger platform (2008). We demonstrate our approach using low-cost, off-the-shelf components including a five degrees of freedom servo manipulator, two CMOS cameras, and a single desktop for all sensing and control tasks. While the performance obtained was not comparable to the Kroger platform, the results are promising given the drastic reduction in cost and resources.
Automatic Deployment and Assembly of Persistent Multi-Robot Formations
With Brian S. Smith, Magnus B. Egerstedt, and Ayanna M. Howard
May, 2007 – December 2008
Currently in Antarctica, automatic weather station (AWS) units are deployed throughout the region for micro-meteorological measurements such as air temperature, wind speed and direction, air pressure, and relative humidity, in order to study the effects of global warming on the Antarctic ice shelves. This work is centered on the design of a distributed network of mobile robots carrying scientific sensors and instruments to augment the existing AWS network. Specifically, this work deals with the issue of automatic deployment to a desired goal location and the assembly of geometric formations (e.g., a hexagon) through the so-called Embedded Graph Grammars (EGGs).
It is assumed that a multiple-robot network consists of agents with a limited sensing range in which they can communicate and estimate the relative position of other network members. The network is also heterogeneous, in that only a subset of robots have global localization ability. Through EGGs, control laws can be automatically generated to ensure that network members without localization are deployed to the correct location in the environment for the sensor placement, and that a rigid and persistent formations is achieved.
Particle Swarm-Assisted State Feedback Control
With Benjamin B. Brackett and Ronald G. Harley
Summer, 2008
The design of feedback controllers to satisfy transient and steady-state performance specifications is well understood in the domain of classical control theory, utilizing frequency-domain approaches such as root locus, Bode, and Nyquist plots in conjunction with simple controllers such as the Proportional-Integral-Derivative (PID) controllers. For time-domain based methods such as state feedback control, however, it is unclear how performance specifications translate into pole locations; i.e., given a set of constraints such as rise time, settling time, overshoot, and steady state error, where the poles should be placed to satisfy these conditions. This work addresses the pole placement problem by using Particle Swarm Optimization (PSO), a stochastic, population-based evolutionary computing technique, to search for the optimal pole locations.
It is demonstrated that PSO works particularly well in locating the optimal poles for Multi-Input Multi-Output (MIMO) systems with a large number of states, where the size of search space is double the size of the state space. Also presented is a proof of concept problem involving the stabilization of a helicopter under the Rationalized Helicopter Model (RHM), and the results of the PSO auto-tuning, including the simultaneous design of a state observer and feedback controller, are supported by illustrative computer simulations.
Survey of Hektor and Oterma through Pulverization of Unique Targets (SHOTPUT)
With Alessondra Springmann, Caley Burke, Megan Cartwright, Rajeev Gadre, Lev Horodyskyj, Andrew Klesh, Keith Milam, Nicholas Moskovitz, Jonathon Oiler, Daniel Ostrowski, Michael Pagano, Ramsey Smith, Shintaro Taniguchi, Amy Townsend-Small, Kongpop U-yen, Steven Vance, Joseph Westlake, and Krzysztof Zacny. Summer, 2008
This work involves the design of a New-Frontiers-class mission to study small bodies in our solar system as a part of the 20th Annual Planetary Science Summer School held in the Jet Propulsion Laboratory, Pasadena, California, USA, from August 4 – 8, 2008. Survey of Hektor and Oterma through Pulverization of Unique Targets (SHOTPUT) is a seven-year mission including a flyby of main belt asteroid (108144) 2001 HM1, a flyby and impactor release at the Trojan (624) Hector (a suspected contact binary) with companion P/2006, and a flyby with impactor release at the Centaur 39P/Oterma. The mission intends to answer questions about the origins, physical properties, and compositions of Trojans and Centaurs and how the color, structure and evolution of these objects compare to those of comets.
SHOTPUT calls for a launch date of March 27, 2015 at the Kennedy Space Center, using the Atlas V531 launch vehicle. The probe is expected to reach the main belt asteroids on June 14, 2016, and flyby Hektor and Oterma on March 25, 2020 and October 30, 2022, respectively. 
The probe carries six instruments:
- Multi-Spectral Imager (Narrow Angle Camera and IR Spec) (MSI)
- Dust Secondary Ion Mass Spectrometer (DSIMS)
- Thermal Infrared Spectrometer (TIR)
- Ultra-Violet Imaging Spectrograph (UVIS)
- Wide Angle Camera (WAC)
- Radio Science Experiment (RSE)
With a total mass of 94kg, total operational power of 98W, and total data rate of 5000 kbit/s. The mission has an expected cost $622.8m.
