Page 32 - Winter 2020

BAT ROBOT EVOLUTION
different directions (Figure 1C; Fazli and Kleeman, 2005). With their size, the sonar rings ushered in the dinosaur
age in the evolution of bat robots.
The sonar rings were used to map out the environment of the robot by associating each reading of the target distance with the direction the respective ranging module was pointed in to fix the location of the target. This approach, however, did little to overcome the many fundamental shortcomings of the sonar range-finding approach outlined above. Because of these difficulties, most robotics engineers abandoned the ultrasonic rangefinders in favor of more powerful techniques such as stereo camera vision or laser range finding (i.e., “light radar” [LIDAR]) as the respective hardware became more widely available and the required computational capacities became more affordable.
Nevertheless, despite improvements to the costs of hardware and especially computing, autonomy solutions based on devices that collect large amounts of data such as laser scanners or stereo cameras remain too costly and too power hungry for applications such as small autonomous drones. If operated under the power and weight constraints of a drone, the processing latencies would be too large to allow for flying through a forest at speed, for example. In this respect, the modalities that replaced the ultrasonic range finders remain stuck in the same evolutionary dead-end road that has led to the extinction of the sonar-ring dinosaurs.
Seeing this situation, some engineers have returned to bats for more insight into how these sensing problems may be solved in a much more elegant and parsimonious fashion. If many millions of bats demonstrate every night that a few grams worth of sensors and computing are sufficient, how difficult could it be for engineers to catch up?
Less Is More: Back to Two Ears
The next step in the evolution of bat robots to lead forward from the ranging modules has focused on smarter ways to combine signals from two receivers. These efforts have been inspired by what bats do with their two ears. Such binaural sensing has been used to improve performance in target tracking as well as in target identification tasks.
The most straightforward way to combine signals from two receivers is to check which receiver detects the input
first. This is similar to the use of binaural time differences in many mammals, including humans, — for telling the direction of a sound source on the horizontal plane. A biomimetic sonar system that used this principle in the horizontal as well as in the vertical employed four transducers implemented in horizontal and vertical pairs to track a target in these dimensions (Figure 2). This bat robot was not only extremely parsimonious in its layout but also with respect to the computations that were required for its operation. All that was necessary was to determine which sensor in the pair detects an echo first (Kuc, 1993). This nonlinear system operated with minimum delay, which is important for optimal sensorimotor operation because the later echo arriving at the contralateral ear was not used.
Bats and humans have two ears instead of four and hence can use time-of-arrival differences only to determine the direction of a sound source in one direction, typically the horizontal plane. In the vertical plane, comparisons of sound amplitude across different frequencies are typically made to determine the direction from which a sound is coming. Such frequency- and direction-dependent effects are particularly easy to model based on the harmonic structure of bat biosonar pulses where each harmonic produces its own beam with a width that decreases with each harmonic. A binaural tracking simulation (Kuc,
Figure 2. Bat robot with two pairs of ultrasonic receivers (circular object on each branch of the cross; “ears”) that are arranged horizontally and vertically. The transducer in the center is the emitter. Pneumatic tubes (with red caps seen only in the horizontal branches) allow the system to track a target in both of these directions based on the sensory inputs.
32 Acoustics Today • Winter 2020