|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Interaction Lab |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Bandit I
A user-friendly humanoid robot torso platform, developed by Living Creatures, details found
here.
Additional images of this platform are found here.
Bandit II
Bandit II is
an updated version of Bandit I, designed and fabricated by BlueSky Robotics. Our lab currently maintains six of these expressive humanoid platforms.
ESRA
ESRA II from Robodyssey Systems LLC.
Nao
The Interaction Lab currently has 11 Nao robots. The Nao robot is a full-body humanoid robot by Aldebaran Robotics. It is approximately 26" tall with 26 degrees of freedom in the neck, shoulders, elbows, hands, hips, knees, wrists, and ankles. It has a number of sensors, including an IMU, sonar, video, microphones, and force sensors. It has two speakers and colored LEDs in the eyes for expressive output.
PR2
The PR2 is a mobile manipulation platform, designed and manufactured by Willow Garage.
Sparky
Minimatronic Figure (TM) courtesy of Walt Disney Imagineering Research and Development.
The Dragonbot is a five degree of freedom dragon robot designed by the MIT Personal Robotics Group. The aesthetics of the robot as seen here were designed by Viva La Puppet.
|
|
|
|
|
Bandit II Simulation
Bandit II humanoid simulation using the ROS visualization environment (RViz)
General Kinematic Simulation
General kinematic humanoid simulator
(developed by Marcelo Kallman)
Humanoid Simulation
Humanoid simulation testbed (using Vortex libraries)
Physic-based Simulation
Physics based humanoid simulation: Adonis (developed by Jessica Hodgins).
The Giraff is a remotely controlled mobile, human-height physical avatar integrated with a videoconferencing system (including a camera, display, speaker and microphone). It is powered by motors that can propel and turn the device in any direction, even backwards. The Giraff can turn in place to face a person or object, and the display can also be tilted up and down to extend the vertical field-of-view.
For experiments in deictic gesture perception and production, we constructed a 12x8 foot freestanding screen with transparent acrylic. The screen is used as a planar target on which we measure both the actual targets of a given deictic gesture and the induced perceived target location by an observer. We have conducted experiments with both robot and human gesturers and have study variables including gesture modality and distance. In addition, we have produced a set of markers of various sizes that can be affixed to the screen to study the effects of saliency on gesture perception. When a participant is placed 3 feet from the screen it covers an horizontal field of view of 120 degrees and a vertical field of view of about 70 degrees.
|
|
|
|
|
|
|
|
Overhead Interaction Toolkit (OIT)
OIT is an overhead 2D visual tracking package intended for tracking humans and robots in indoor settings. OIT will work with any camera compatible with the ROS project. Research involving the OIT concerns detecting and interpreting the behavior of people as they interact with robots.
Development of the OIT was done in C++ and python, but is compatible with any language used by the ROS project. The OIT was developed on x86 hardware, but probably is compatible with any POSIX-compliant OS and hardware.
A wireless motion tracking system consisting of 3-axis IMUs that measure acceleration, angular change of rate, and magnetic field strength. This system uses an onboard wearable computer to store, process, and transmit these data to a local laptop or PC, and is an updated version of the motion suit developed previously in the lab.
This system is currently being used in a robot-based rehabilitation framework. The data from the sensors can be used to assess quality of motion as participants interact with the robot and perform various motor tasks.
In the past, the system has also been used for other projects, including teleoperation of the NASA Robonaut.
An overhead vision system for tracking our robots and other lab activities (created by Paolo Pirjanian).
A motion tracking system for recording human torso movement (created by Stefan Weber).
Mezzanine is an overhead 2D visual tracking package intended primarily for use as a mobile robot metrology system. It uses a camera to track objects marked with color-coded fiducials and infers the pose (position and orientation) of these objects in world coordinates. Mezzanine will work with most color cameras (even the really cheap ones) and can correct for the barrel distortion produced by wide-angle lenses.
Mezzanine is language and architecture neutral: object poses are written into a shared memory-mapped file which other programs can read and make use of as they see fit. Mezzanine itself is written in pure C and was developed under Linux on x86 hardware.
Laser-based motion-tracking system
One of our research directions involves activity tracking and modeling. We use multiple planar laser range finders to track robots and people in the lab. The lasers are automatically calibrated relative to each other using mesh relaxation. A background model representing the positions of static objects is maintained. Range readings that do not fit this model are assumed to come from moving object. Foreground points are then clustered into groups representing the boundaries of objects to be tracked. We have used two approaches for tracking targets:
The figure (click for a larger image) shows a snapshot of 3 lasers tracking two people walking in the lab. The people are the two clusters of particles, the blue edges are the background model.
For more information on our laser tracking projects, visit Helen Yan's webpage.
Volume-based markerless motion capture is a motion tracking system. It estimates both kinematics model and joint angles from volume data extracted by multiple cameras.
Our constantly growing collection of robots started with 20 R1s (affectionately dubbed the Nerd Herd and named after things that come out of a toaster; now retired) and 4 R2s (affectionately dubbed the Socially Mobile and named Don Quixote, Don Corleone, Donna E. Mobile, The Donald and Don' Work, now also retired). The R1s and the R2s were generously given to the Interaction Lab by Prof. Rodney A. Brooks from the MIT AI Lab. We then added two Pioneers to the family, Benjamin and Mae (the little Volens), then a few more indoor and outdoor versions, and are now, as part of the USC Robotics Lab, up to to 13 Pioneer robots (9 Pioneer 2 DX, 2 Pioneer 1s, and 2 Pioneer 1 ATs), and continue to grow that part of the robot family. We also have two Sony Dog Robots we call Laika and Dogmatix, four tiny Khepera robots, a lawn-mowing robot, and a Phoenix robot from Johuco. Evolution Robotics has also provided us with six ER1 robots, one ER1 SDK, and one Scorpion.
 Robot Family (2000) |
 Robot Family (2003) |
|
|
| Three Evolution Robotics mobile robots. From left to right: ER1, ER1 SDK, ER Scorpion |
Player is a multi-threaded robot device server. It gives you simple and complete control over the physical sensors and actuators on your mobile robot (Player is currently specifically implemented for the ActivMedia Pioneer 2-DX mobile robot running Linux, but the extensible server architecture should facilitate porting to other platforms). When Player is running on your robot, your client control program connects to it via a standard TCP socket, and communication is accomplished by the sending and receiving of some of a small set of simple messages. In addition to talking to your hardware, Player can interface to the robot simulator Stage.
Stage simulates mobile robots moving in and sensing a two-dimensional bitmapped environment, controlled through Player. Player provides a powerful, flexible interface to ActivMedia Pioneer 2 robots; Stage provides virtual Player robots. Various sensor models are provided, including sonar, scanning laser rangefinder, pan-tilt-zoom camera with color blob detection and odometry. Several controllers designed in Stage have been demonstrated to work on real robots.
Gazebo is a mobile robot simulator much like Stage with a focus on outdoor environments. Like Stage, it is capable of simulating a population of robots, sensors and objects, but does so in a three-dimensional world. It generates both realistic sensor feedback and physically plausible interactions between objects (it includes an accurate simulation of rigid-body physics).
For more information, please see this page.
The library includes conference proceedings, journals, and other reading material avilable in the lab.
