Bats could help the development of AI robots

Rolf Müller – rolf.mueller@vt.edu
X (twitter): @UBDVTLab
Instagram: @ubdvtcenter
Department of Mechanical Engineering, Virginia Tech, Blacksburg, Virginia, 24061, United States

Popular version of 4aAB7 – Of bats and robots
Presented at the 186th ASA Meeting
Read the abstract at https://doi.org/10.1121/10.0027373

–The research described in this Acoustics Lay Language Paper may not have yet been peer reviewed–

Given the ongoing revolution in AI, it may appear that all humanity can do now is wait for AI-powered robots to take over the world. However, while stringing together eloquently worded sentences is certainly impressive, AI is still far from dealing with many of the complexities of the real world. Besides serving the sinister goal of world-domination, robots that have the intelligence to accomplish demanding missions in complex environments could transform humanity’s ability to deal with fundamental key challenges to its survival, e.g., production of food and regrowable materials as well as maintaining healthy ecosystems.

To accomplish the goal of having a robot operate autonomously in complex real-world environments, a variety of methods have been developed – typically with mixed results at best. At the basis of these methods are usually two related concepts: The creation of a model for the geometry of an environment and the use of deterministic templates to identify objects. However, both approaches have already proven to be limited in their applicability, reliability, as well as due to their often prohibitively high computational cost.

Bats navigating dense vegetation – such as in rainforests of Southeast Asia, where our fieldwork is being carried out – may provide a promising alternative to the current approaches: The animals sense their environments through a small number of brief echoes to ultrasonic pulses. The comparatively large wavelengths of these pulses (millimeter to centimeter) combined with the fact that the ears of the bats fall not too far above from these wavelengths on the size scale condemns bat biosonar to poor angular resolution. This prevents the animals from resolving densely packed scatterers such as leave in a foliage. Hence, the echoes that bats navigating under such conditions have to deal with inputs that can be classified as “clutter”, i.e., signals that consists of contributions from many unresolvable scatterers that must be treated as random due to lack of knowledge. The nature of the clutter echoes makes it unlikely that bats having to deal with complex environments rely heavily on three-dimensional models of their surroundings and deterministic templates.

Hence, bats must have evolved sensing paradigms to ensure that the clutter echoes contain the relevant sensory information and that this information can be extracted. Coupling between sensing and actuation could very well play a critical role in this. Hence, robotics might be of pivotal importance in replicating the skills of bats in sensing and navigating their environments. Similarly, the deep-learning revolution could bring a previously unavailable ability to extract complex patterns from data to bear on the problem of extracting insight from clutter echoes. Taken together, insights from these approaches could lead to novel acoustics-based paradigms for obtaining relevant sensory information on complex environment in a direct and highly parsimonious manner. These approaches could then enable autonomous robots that can learn to navigate new environments in a fast and highly efficient manner and transform the use of autonomous systems in outdoor tasks.

Biomimetic robots designed to reproduce the (a) biosonar sensing and (b) flapping-flight capabilities of bats. Design renderings by Zhengsheng Lu (a) and Adam Carmody (b).

As pilot demonstration for this approach, we present a twin pair of bioinspired robots, one to mimic the biosonar sensing abilities of bats and the other to mimic the flapping flight of the animals. The biosonar robot has been used successfully to identify locations and find passageways in complex, natural environments. To accomplish this, the biomimetic sonar has been integrated with deep-learning analysis of clutter echoes. The flapping-flight line of biomimetic robots has just started to reproduce some of the many degrees of freedom in the wing kinematics of bats. Ultimately, the two robots are to be integrated into a single system to investigate the coupling of biosonar sensing and flight.