Current Research Projects


Brain Infrastructure for Coco

Charlie Kemp

coco front view
link to ps file link to pdf file
We wish to use Coco, a small ape-like robot shown below, to explore humanoid intelligence. Sophisticated humanoid robots such as Coco require large amounts of parallel computation in order to support rich real-time multi-modal sensory input and complex behavioral output. Creating a computational system of this nature is difficult.

A well-designed computational infrastructure will help us to organize the requisite computation and allow us to more easily add functionality to Coco. For example, along with other code we would like Coco's brain to simultaneously run a variety of real-time vision algorithms, audio processing algorithms, tactile information processing algorithms and motor control algorithms. Without a framework for making these algorithms work together the effort spent maintaining the system, let alone producing new research at the creature level, would be infeasible.

[Back to Top]


current research button
coco side view
Building Affective Robots

Juan D. Velasquez

link to ps file link to pdf file
Affect is inherently intertwined with several attributes that we associate with intelligent behavior, such as multimodal sensory-motor integration, attention, natural social interactions, and learning and development. The primary focus of this research is to investigate affect from a computational perspective by implementing models of affective processing on a variety of robotic platforms.

Recently, efforts have shifted from behavior-based approaches to robotics that deal with insect-level competence, to those that try to build humanoid robots with an increasingly complex behavioral repertoire that includes, among other competences, the ability to interact socially.

If we are to build robots that exhibit robust and adaptive behavior, and which are responsive to social interactions, then we must deal with issues such as motivation and emotion, which have not been considered essential in previous behavior-based robotics, but which are crucial in humanoid robotics.

[Back to Top]


Medial Axis Representations for the Analysis of Human Shape in Images

Charlie Kemp

link to ps file link to pdf file
Since the 1960's researchers have observed that biological shape can be usefully described in terms of medial axes. Unfortunately, the traditional algorithms used to estimate medial axis descriptions of shape are of limited use due to a variety of practical problems, including instability to minor perturbations of the input.

By using more recent signal processing techniques in conjunction with an updated representational framework, the benefits of medial axis representations can be exploited while avoiding several of their traditional drawbacks. Some types of medial axis representations appear to be well matched to the class of shapes that comprise human faces and bodies. These descriptors highlight the underlying structure of the overall shape, while also representing important multi-scale aspects of the component shapes. As an intuitive example, a stick figure representation of a human body does a good job of encoding the body's configuration with respect to the world and the observer. Furthermore, each part of the stick figure can describe a component shape of the body at some scale, such as an elongated rectangular shape for the torso and legs, and two smaller tapering tubes for the legs.

[Back to Top]


A Mobile Robot for Social Interaction

Christopher Morse

Coco
link to ps file link to pdf file
To create a mobile robot with morphology that is conducive to human interaction and equipped with sufficient sensors and actuators that social behavior is possible. This research is a continuation of the work done on the Yuppy robot.

Artificial social behavior models require a platform that provides both sensory input and a medium for behavioral output. The amount and quality of sensory information provided to the software directly impacts performance and capabilities for further development. In the same way, the repertoire of expression directly effects how well the system will be perceived to be working. A mobile platform adds an entirely new dimension to both the sensory systems and behavior generation by allowing everything the robot senses and does to be related to physical position. Human factors also play into the success of a particular behavior model; if people have difficulty interacting with the platform then the abilities of the model will be put at a disadvantage.

[Back to Top]


Motor Control System for a Quadruped Robot

Eduardo Torres-Jara

coco walking
link to ps file link to pdf file
To develop a Motor System for a Quadruped Robot that allows it to control its gestures, posture and walking. In order to obtain any useful behavior from a robot, it is necessary to control the movements of its limbs. The movements of the limbs should be smooth and adequate to the task (i.e., if the robot is standing, it would control the force exerted by the limbs against a surface, and if the robot is pointing, it would control the position of its arm.)

[Back to Top]


webmaster: annika