Sonar Sensor Interpretation and Infrared Image Fusion for Mobile Robotics

The goal of our research is to give mobile robots the ability to deal with real world instructions outdoors such as “Go down to the big tree and turn left.” Inherent in such a paradigm is the robot being able to recognize a “big tree” and do it in unstructured environments without too many pre-mapped fixed landmarks. Ultimately we envision mobile robots that unobtrusively mix in with pedestrian traffic and hence traveling primarily at a walking pace. With regard to sensors this is a different problem from robots designed to drive on roadways, since the necessary range of sensors is tied to the speed of the robot. It’s also important to note that small mobile robots that are intended to mix with pedestrian traffic must normally travel at the same speed as the pedestrians, even if they occasionally scurry quickly down a deserted alley or slow way down to traverse a tricky obstacle, because people resent having to go around a slow robot while they are also easily startled by machines such as Segways that overtake them without warning. At walking speeds the range of sonar at about 50kHz is optimal, and there are none of the safety concerns one might have with lidar, for example. This type of sonar is precisely what bats use for echolocation; the goal of our research is to employ sonar sensors to allow mobile robots to recognize objects in the everyday environment based on simple signal processing algorithms tied to the physics of how the sonar backscatters from various objects. Our primary interest is for those sensors that can function well outdoors without regard to lighting conditions or even in the absence of daylight. We have built several 3D sonar scanning systems packaged as sensor heads on mobile robots, so that we are able to traverse the local environment and easily acquire 3D sonar scans of typical objects and structures. Of course sonar of the type we’re exploring is not new. As early as 1773 it was observed that bats could fly freely in a dark room and pointed out that hearing was an important component of bats’ orientation and obstacle avoidance capabilities (Au, 1993). By 1912 it was suggested that bats use sounds inaudible to humans to detect objects (Maxim, 1912) but it wasn’t until 1938 that Griffin proved that bats use ultrasound to detect objects (Griffin, 1958). More recently, Roman Kuc and his co-workers (Barshan & Kuc, 1992; Kleeman & Kuc, 1995) developed a series of active wide-beam sonar systems that mimic the sensor configuration of echolocating bats, which can distinguish planes, corners and edges necessary to navigate indoors. Catherine Wykes and her colleagues (Chou & Wykes, 1999) have built a prototype integrated ultrasonic/vision sensor that uses an off-the-shelf CCD camera and a fourelement sonar phased array sensor that enables the camera to be calibrated using data from