Schools of Fish School Robots

Swarm robotics holds significant potential for efficiently handling large-scale, distributed tasks. This approach leverages multiple simple, affordable robots working together to achieve robust and scalable results. By collaborating as a team, these robots can tackle complex jobs in a variety of fields, such as agriculture, environmental monitoring, and search and rescue. Through this collective approach, swarm robotics can accomplish tasks that would be impossible for a single robot to achieve alone.

But it’s not all sunshine and rainbows. Developing effective control algorithms is fraught with difficulties. Ensuring that the swarm operates cohesively, adapts to changing environments, and maintains stability in the face of individual robot failures are important considerations that researchers and engineers must address to fully realize the potential of swarm robotics. However, these issues have not yet been adequately dealt with.

A group led by researchers at the University of Konstanz in Germany has decided to approach this problem from the perspective of a fish. Many types of fish instinctively swim in schools (i.e., swarms) to evade predators, hunt for food, or improve their swimming efficiency. So the team reasoned that they should be good teachers for robots that are trying to find their way in a sea of fellow machines.

To uncover the hidden rules of fish schooling, the researchers built a cutting-edge virtual reality system that allows real zebrafish to interact with holographic virtual fish. Each virtual fish was actually a projection of a real fish from another tank, networked into a shared 3D digital environment. This setup let the researchers precisely manipulate visual cues and isolate the specific sensory-motor responses that drive fish coordination.

What they discovered is that zebrafish do not need to know how fast their neighbors are moving, or much of anything else, aside from where they are. The fish rely almost entirely on the perceived position of nearby individuals to guide their movement. Based on this, the team designed a simple control rule — a kind of proportional derivative (PD) controller they named BioPD — that accurately captures how real fish pursue and respond to their peers.

To test the realism of their algorithm, the team performed a sort of aquatic Turing test. A real fish was introduced to swim alongside a virtual partner, which alternated between mimicking a real fish and following the BioPD algorithm. The fish responded identically in both scenarios, suggesting that the virtual follower governed by BioPD was indistinguishable from a biological one.

The researchers then tested BioPD on land, in the air, and at sea by integrating it into robotic cars, drones, and boats. In all cases, the BioPD-powered robots performed comparably to traditional high-complexity control algorithms, like Model Predictive Control, but with much lower computational demands. This success was due to the fact that the algorithm is based on simple kinematic rules that work across platforms without significant customization.

This research not only sheds light on the elegant simplicity of fish schooling, but is also a meaningful step forward in the design of low-complexity, high-efficiency swarm robotics control systems. By looking to nature for inspiration, the team has offered a biologically grounded blueprint for future robot coordination strategies.