The goal of my work is to bring computer vision and robotics to the common man by creating systems that make the task of driving easier and safer. A key system requirement in the domain of driving is the ability to adapt to changing conditions, since the appearance of the car's surroundings can change dramatically depending on environmental conditions and the type of road the car is on. My work and that of my students focuses on the development of adaptive techniques to achieve this flexibility.
Collision Warning/Avoidance Systems: Over forty thousand people die on US roads every year in traffic accidents. One of my goals is to save some of these lives through the development of systems that warn the driver of impending collisions, and potentially take momentary control of the vehicle to prevent a crash. My students and I are currently focused on the problem of single vehicle roadway departure crashes, in which the vehicle drifts off the road because the driver isn't paying attention, has fallen asleep, or is in some way impaired (intoxicated, medical emergency, etc.). Using a video camera to monitor the road ahead, in conjunction with digital maps, we are developing prototype systems on the Navlab test vehicle that sound an audible tone when the vehicle begins to drift from its lane. If the driver doesn't respond to the warning, the system can assume momentary control of the steering wheel to return the vehicle to the roadway. Critical research issues associated with the problem include reliable perception of the vehicle's surroundings to minimize false alarms, adaptive warning algorithms to accommodate different driving styles, and man-machine interfaces to effectively alert the driver and avoid the crash.
Automated Highway System (AHS): A more futuristic effort my students and I are involved in is the Automated Highway System (AHS). The ambitious goal of this project is to improve safety and reduce congestion on our nations highways by fully automating the driving task. A key research issue in this domain is reliable perception of the road, other vehicles and obstacles. We are investigating solutions to this problem using various image processing techniques, as well as alternative sensing modalities like infrared lasers and millimeter wave radar. Another important area we are addressing is the planning and coordination of vehicle maneuvers such as lane changes and obstacle avoidance. Finally, we are developing control techniques that allow the vehicle to actually execute these maneuvers.
Using the results of our research, we have demonstrated systems capable of autonomously driving the Navlab vehicle for long distances at speeds of up to 90 mph. One such vision system, called RALPH (Rapidly Adapting Lateral Position Handler) steered the Navlab for 98.2% of a trip from Washington, DC to San Diego, CA, a distance of over 2800 miles. When augmented with a millimeter wave radar for sensing obstacles, the system can detect vehicles ahead and automatically change lanes when appropriate.
|Research Interest Keywords|
|computer vision, machine learning, mobile robots, obstacle avoidance, sensors|
|The Robotics Institute is part of the School of Computer Science, Carnegie Mellon University.|
Contact Us | Update Instructions