Research stage: early to middle

Human responses to emergency situations such as dangerous chemical spills, severe weather, or escalating political crises, are often negatively impacted by not enough correct information in the early response stages, and the daunting task of rapidly coordinating disparate and distributed emergency responders, each with partial and possibly incorrect knowledge. This presentation illustrates how context-aware autonomous agents can help emergency responders to more quickly assess the state of the crisis, and assist in the coordination of their responses. The solution is unique due to the maturity of the underlying software architecture (RETSINA). We are the only group in the US applying autonomous software agents to HazMat response and evacuation operations.

Computational mechanics simulations of deformable objects and mechanical assemblies are notoriously expensive when the size and/or physical complexity of the system is high. These costs are particularly prohibitive for realtime applications such as immersive virtual environments, haptic rendering, or surgical simulation. This talk provides an overview of my research group's work on data-driven pre-computation methods. After modest preprocessing, these emerging technologies can provide enormous speedups for nonlinear deformation computations, collision processing, global illumination, and related multimodal interaction applications.

We have developed algorithms for flexible, robust, and human controllable coordination between hundreds of thousands of highly heterogeneous team members.  By leveraging the power of a carefully designed logical network connecting team members, coordination reasoning is spread around the network maximizing reactiveness while minimizing communication overhead.

We have developed a realtime system for tracking a person's head and estimating the head position (xyz), head pose (3 angles), and gaze (2 angles). Our system uses a single camera and is more accurate than other systems.

We have developed a complete system for building 3D articulated human body models, tracking a person with those models, and re-rendering the person in new postures. Our system is entirely camera-based and requires no markers. Neither laser scanners nor dedicated motion capture systems (with their associated markers) are required.

Snake robots are highly articulated devices that can poke through tightly packed volumes accessing locations that people and machinery otherwise cannot. We also talk about our mobile robot applications.

The Carnegie Mellon University's NanoRobotics Laboratory has been developing new polymer fiber adhesives inspired by gecko foot-hairs that could stick to almost any material in any environment.  They are self-cleaning, and could be also easily peeled off.  They have a wide range of applications for novel surface-climbing robots, space tape, medical adhesives, sports, entertainment, etc.

A new invention at the Robotics Institute, the Sonic Flashlight^TM, enables doctors to see where they are inserting a needle or a scalpel, by superimposing an ultrasound image on the direct line of site in real time.  Extensions of this technology to non-medical technology will allow fire-fighters to see through smoke and scuba divers to look through murky water.

We are developing systems and algorithms for automated video surveillance and biometric identification. We have demonstrated realtime detection, tracking, and classification by an outdoor multi-camera surveillance system. We have developed practical algorithms for human identification using face and gait biometrics. We are currently developing computer vision algorithms that observe people in the scene to learn what activities are being performed, and who typically performs which actions. These milestones are significant steps towards our long-term goal of building intelligent systems that understand the realm of human activity well enough to aid and interact with people.

This talk outlines analysis, prototyping, and management tools intended to improve mobile and stationary sensor network selection and performance.  The described work is distinguished by a comprehensive approach including mission management, task decomposition, and sensor scheduling based on both time constraints and the notion of sensor "suitability".

Sensor networks are composed of small, inexpensive, wireless nodes that can sense the environment and control devices.  They take advantage of recent advances in computing and wireless communication. We expect a billion-dollar market for sensor networks by 2010 with applications in agriculture, security, defense, and home automation.  We have developed a robust, low-cost system that is in continuous use in a farm. I will discuss the status of our project as well as potential applications.

Military users and first responder's lack a reliable means of tracking an individual's position in areas where usage of Global Positioning System (GPS) or other infrastructure is denied. Examples of these "GPS-denied" areas would be in urban canyons, operating inside of buildings, or walking under canopy or underground. We present a concept, called Personal Sensor Webs (PSW), and results obtained from two generations of PSW position tracking technology.

As technology becomes more and more complex, it becomes harder and harder to interact with. Autonomous mobile robots are one extreme example of that, but many other common appliances (such as VCRs, microwaves, computers) are at, or near, the point of being inscrutable to the average user. One approach is to make the technology conform to how people interact, rather than the other way around. We are developing techniques for human-robot social interaction, in which the robot models the intentions of the user in order to interact in a socially acceptable manner. This includes issues of free-form dialogue, mixed initiative interaction, interfaces with socially acceptable personalities, and expressive interfaces.

An Intelligent (sheet metal) Bending Workstation starts with a design for a part and plans the sequence of bending operations, the layout of the tooling stages, the robot gripper's grasping locations, and the robot motions and bending machine operations required to make the part. Our approach is unique in that we automatically generate bending process plans using geometric reasoning, fast search algorithms, and constraints of the bending machine and tools while the conventional planning is manually done by modifying the plans designed for the similar parts. Our system generates plans in less than a minute for most parts while conventional planning could take from tens of minutes to a few days. Our system can also be used as a part design tool because it can tell whether the part can be manufacturable at a job shop.

Given a large set of digitized biomedical images ranging from microscope films to CT or MR scans and their collateral information, we develop a learning-based system to identify a discriminative image feature subspace for the semantic characterization of image similarity. Our approach is unique in that it starts from raw image data, explores a huge image feature space, and effectively reduces the space down to a discriminative subspace with quantifiable evaluations.

NYA

We will present a mobile vehicle automation technology that enables semi-autonomous operation of mobile work vehicles, such as those employed in agriculture, mining, and construction.  Such automation promises to improve productivity, reduce operational costs, and improve safety.

There is strong demand throughout the educational system for technology and curricula that can stem the tide of losses in technology literacy and the falling enrollments in engineering and technological science fields.  We have a set of innovative technologies that have been successfully piloted across the US with great success.

Involuntary hand tremor is naturally present in even the most skilled eye surgeons. The Micron device actively cancels tremorous motions using novel sensing, signal processing, adaptive filtering, and miniature robotic mechanisms, all in a single handheld form factor.

Many therapies for heart disease require the surgeon to access the outer surface of the heart (also known as the epicardium). Gaining that access, even with “minimally invasive” techniques, generally necessitates three additional procedures unrelated to the actual cardiac intervention: deflating a lung, placing the patient on ventilation assistance, and use of general anesthesia. An alternative approach that avoids those additional procedures is HeartLander - a small mobile robot that traverses the epicardium to the location desired by the surgeon.

Emerging guidance and control systems will be outlined which reduce or eliminate the traditional reliance of automated guided vehicles surveyed in-plant infrastructure. Techniques for achieving more seamless integration of manned and automated industrial trucks and enhancing the safety of both will be presented.

The National Robotics Engineering Consortium (NREC) of the Robotics Institute is driven to move robotics technologies from the lab to the commercial marketplace. This talk will highlight, through the presentation of case studies, NREC project management processes that support rapid development while reducing risk.

The TechBridgeWorld initiative was created in 2003 with the goal of investigating the role technology can play in enabling and enhancing sustainable development around the globe.  The different programs under this initiative aim to encourage relevant discourse and understanding within the academic community of the positive intersection between technology and development, and of the importance of training technology-aware citizens, and civic-minded technologists, as future leaders.

Information is growing exponentially and all future generations of children will be required to compete globally. It is imperative that adults make it easier for children to learn to compete. Carnegie Mellon's Robotics Academy  is a leading developer of educational tools designed to excite and motivate this generation of children to pursue degrees in science and technology.

Imagine twenty robotic vehicles racing across the desert to Las Vegas. Sounds like a sequel to Mad Max the movie, but it happened on March 13, 2004 and it will happen again in October 2005. This was DARPA's Grand Challenge to build a fully autonomous offroad vehicle.

We will show video images of the race and discuss the technological breakthroughs that went into Sandstorm and the challenges that remain. You'll also be the first to hear about Highlander; Carnegie Mellon's latest entry in the 2005 Grand Challenge race.

The autonomous vehicles in the Grand Challenge foreshadow future robotic vehicles that will explore deep space and perform heroic missions such as repairing the Hubble space telescope.

This talk covers a variety of topics including technology licensing and intellectual property aspects of sponsored research agreements.

Universities operate differently than corporations.  This talk explains the difference and details how a sponsor can get the most out of a research project while avoiding the pitfalls.

Founded in April 2003, the Robotics Foundry is an independent, non-profit economic development organization that directs programs and initiatives to accelerate the growth of Agile Robotics Technologies and to establish a significant industry cluster in western Pennsylvania.