| Cornell B.R.A.I.N. Team: |
Autonomous Underwater Vehicles (AUVs) are an exciting topic for two reasons. First, there are many interesting real-world applications for such systems. Effective autonomous underwater vehicles would allow or facilitate exploration, salvage, search and rescue, and scientific studies in deep ocean areas. Second, the highly dynamic and noisy nature of the underwater environment makes the problem a difficult one. Combined with the noise and dynamics of the environment are the additional problems of a possible lack of reference points and limited communications due to the water itself. My research in underwater robotics has been as a member of Cornell University's B.R.A.I.N. (Big Red Artificial Intelligence Navigator) team for two years. My main interests in this area are in high-level AI and robotic control. I have looked at and implemented some solutions for the problems inherent in underwater robotic control. These include: path planning and control in a highly dynamic environment in the context of a free floating differential thrust submarine and submarine localization with high sensor noise and sparse landmarks. The problem of localization is particularly interesting and can be stated as: "How can the AUV know where it is?" This information is critical in order for the AUV to accomplish even the basic actions, such as moving to another point. In land-based systems, this might be accomplished using GPS or landmark based localization. For the AUV competition, GPS was not permitted (regardless of whether is might have worked underwater). Landmarks are sparse underwater and generally confined to features on the floor. Further, all of the sensor data was shown to contain high amounts of noise. This makes the problem: "How can a robot determine where it is in three dimensional space from sparse and inaccurate sensor information that it has been receiving for a given amount of time, namely since the robot began to move?" Our approach to solving this problem is a simplified version of the Monte Carlo Localization work done at Carnegie-Mellon.  |
