You are here: Foswiki>RoboticsWeb Web>AvailableProjects (30 Aug 2013, trink@LAB.CS.RPI.EDU)EditAttach

Project Areas

The projects listed below are of central importance to the research of our lab. However, you may propose a project not listed here, combine parts of several projects into a new one, or extend projects in new directions that suit your interests. The projects have needs for students with interests in any of the following: database systems, real-time systems, dynamics of linkages, control theory, image processing, machine learning, hidden markov models, particle filtering, optimization, system identification, data mining, and biomechanics.

Grasp Acqusition - How do people and robots pick things up?
Grasp-SLAM - Estimation and modeling of the grasp acquisition process.
Fast Accurate Physics Engines - Leveraging parallel computing hardware for super-fast physics simulation.
Geometric Problems - Develop better geometric models and fast distance computation algorithms.
High Performance Control - Improve the performance of our robot control systems.
Mobility Analysis - How could robots move more efficiently over rough terrain.
Haptic Feedback - Develop a haptic interface to a simulated or real robot that can feel interaction forces.

Grasp Acquisition (GA) Projects

Humans are really good at grasping things. Robots are not. What's hardest for robots is the transition from a configuration where there is no contact between the hand and the object to one in which the hand grasps the object stable (and even harder, in a way which facilitates a subsequent manipulation task). The theme of the GA projects is to understand how to make robots as reliable and fast as humans at grasping.

GA1 Design and conduct a set of controlled robot grasp acquisition experiments using the Barrett Hand/Arm system. Gather motion-capture and video data. Store it in the Grasp Acquisition Database.
GA2 Develop a simulator for of the Barrett and/Arm system. Identify the parameters of the simulation model to match the experiments stored in the Grasp Acquisition Database as closely as possible.
GA3 Design and conduct a set of controlled human grasp acquisition experiments with and without compromised tactile or visual feedback. Gather motion-capture and video data. Store it in the Grasp Acquisition Database.
GA4 Data mine the grasping trajectories from project GA3 looking for features common to successful grasps. What features do successful human grasp have in common? ...robot grasps? Are any of the success indicators common to both human and robot grasping? Can you identify human strategies and the situations under which certain ones will be employed?
GA5 Develop a post-processing tool to segment the data gathered in projects GA1 and GA3 into intervals corresponding to individual phases of the grasping process. This project will involve the use of dynamic Bayesian networks.
GA6 Build a hand with "sticky" fingertips and control algorithms to enhance grasp acquisition. Papers by Mark Cutkosky of Stanford and Ron Fearing of Berkeley which describe the use of sticky surfaces in climbing robots, can serve as a starting point.

G-SL(AM)^2 Projects

Human grasping benefits from our ability to mentally model the object and to use past experience to predict its motions while the fingers are closing. With knowledge of the shape, mass distribution, and surface texture, we are quite adept at picking up and manipulating objects. This project is one of the first in the world to study the G-SL(AM)^2 problem (Grasping - Simultaneous Localization And Modeling And Manipulation). Experiments will be conducted with the Planar Grasp Testbed and the Barrett Arm/Hand system.

G-SLAM1 Improve the tactile sensors in the Planar Grasp Testbed.
G-SLAM2 Augment the Planar Grasp Testbed with a second pusher. Implement synchronized mechanical impedance controllers for the pushers with the ability to change the impedance properties in one millisecond or less. Experimentally verify that the desired impedance behavior is achieved.
G-SLAM3 Develop a simulation tool for the Planar Grasp Testbed that can simulate all of the real experiments that can be done with the Testbed.
G-SLAM4 Use the simulation model from project G-SLAM3 to plan reactive grasp acquisition strategies that are robust to errors in the initial pose of the object.
G-SLAM5 Develop a real-time method for tracking the object's motion and estimating parameters of the simulation model.
G-SLAM6 3D grasping. Work on the topics described in G-SLAM3, 4, or 5, on the Barrett Hand/Arm System.

Fast Accurate Physics Engine (FAPE) Projects

Our goal here is super-real-time, accurate, physics engines for systems experiencing intermittent contact. Such simulations will make it possible to develop reactive controllers for robots to reliably perform dull, dirty, and dangerous jobs, such as complex assembly tasks deep underwater. They will also improve the capabilities of smart prosthetic hands, arms, and legs, increase the speed and accuracy of computer-aided design of machines, and the physical realism of special effects in movies and computer games. Our goal is to increase simulation speed by three orders of magnitude for highly complex scenes by exploiting modern parallel computing hardware including RPI's CCNI and GPU systems.

FAPE1 Develop benchmark simulations related to robots running on granular terrain in Blender. Execute the benchmarks using dVC3d and Solfec and compare the results.
FAPE2 Design and perform physical experiments to measure the traction of robots running on granular terrain.
FAPE3 Design data structures and algorithms for computing the contact and proximity information from bodies represented as polyhedral meshes.
FAPE4 Explore the use of R-functions for representing bodies and computing contact and proximity information.
FAPE5 Develop a method for stochastic characterization of traction of a robot foot on granular terrain.
FAPE6 Compare simulated and physical traction experiments through their stochastic traction characterizations.

Geometric Problems (GP) Projects

A robot's manipulation skills are limited by the quality of geometric models of the interacting bodies (object, fingers, table top, etc). The goal of this problem set is to develop representations and algorithms for geometric models that are suitable for very fast accurate simulation and real-time control.

GP1 Develop a method to construct efficient R-function (paper by Pasko and Adzhiev) representations of objects using data from a Kinect.

High-Performance Control (HPC) Projects

A robot's manipulation skills are limited by the quality of control of the motions of the arm and hand. This quality is impacted by the accuracy of the dynamic and kinematic models of the system and the control method implemented. The objective of the HPC projects is to improve the accuracy and speed of trajectory following, to develop impedance controllers, and to coordinate the motions of the fingers with the arm. These capabilities will enhance our ability to perform the grasping research described above.

HPC1 Port Prof Schaal's (USC) controller to our Barrett Hand/ARM system. Compare it's performance to that of our Matlab controller.
HPC2 Develop a joint friction compensator for the Barrett Hand/ARM system. The compensator could be implemented as a torque control loop that ensures that the torque commanded by the controller is felt at the output by adding more torque to cancel joint friction.
HPC3 Develop a method for automatically calibrating the kinematic and dynamic model of the Barrett Hand/ARM system. This method should also allow the model to be updated to reflect an object grasped by the hand or released from the hand. Compare accuracy with kinematic model provided by Barrett.
HPC4 Develop an operational-space impedance controller for the Barrett Hand/ARM system. Verify that the desired impedance is accurately implemented.
HPC5 The G-SLAM2 project above is the same idea as HPC4, but will be implemented in the simpler Planar Grasp Testbed.

Mobility Analysis (MA) Projects

Animals are capable of ambulating efficiently across many types of terrain. Robots are not, but we'd like them to be.

MA1 Develop a simulation of a mobile robot that can run or drive over piles of loose stones. Systematically study vehicle mobility as a function of number of stones, stone size distribution, friction coefficient, and slope of the underlying support surface.

Haptic Feedback (HF) Projects

Haptic devices give the opportunity to include a person and therefore human perception and problem-solving abilities in the control loop of a complex system. Projects here are designed to develop haptic capabilities for our lab.

HF1 Revive our Phantom 1 haptic device. The mechanical design is fantastic, but we need to scrap the original controller hardware and build a new one. A good design to start with can be obtained from the GRASP Lab at University of Pennsylvania, where they have many revived Phantoms.
HF2 Develop a real-time interface to our Barrett Hand/Arm system and to a physics engine.

-- JeffTrinkle - 2011-08-28
-- LiZhang - 2010-07-26
Topic revision: r12 - 30 Aug 2013, trink@LAB.CS.RPI.EDU

This site is powered by FoswikiCopyright © by the contributing authors. All material on this collaboration platform is the property of the contributing authors.
Ideas, requests, problems regarding Foswiki? Send feedback