Robots learning animal language
Extensive observational studies have produced fundamental information on communication among animals. Social interaction in a group is not only responsible for collective behaviours but also drives the self-organisation of the group. Generating biomimetic robots The EU-FET Proactive ASSISI_bf project developed robots capable of influencing the collective behaviour of animals, focusing on honeybees and fish, known for their collective swarm intelligence. “Our main goal was to establish a robotic society capable of communicating with animal societies and learning from them,″ explains project coordinator Dr Thomas Schmickl. Scientists had to address major challenges in understanding how animals behave as groups. Following years of pure biological experimentation and mathematical modelling, they found a way to successfully decipher these collective decision-making systems. The team generated ASSISI_bf autonomous robots using evolutionary algorithms that allowed them to adapt to the animal swarms and learn to interact with animals in a desired way. These long-lasting robots were designed with precise biomimetic behaviour to ensure acceptance by the animals. Honeybee robots were static but required extensive testing with regards to temperature production, vibration and production of a subtle air flow that resembled the beating of wings. For this purpose, researchers utilised applied machine learning and evolutionary computation techniques. Fish robots, on the other hand, were generated to move around like real fish. To achieve that, scientists coupled the robot with an actuated robotic fish lure capable of beating its tail to increase its attractiveness among fish. Using 3D printing based on scans of real animals, robots were covered with a decal to have the same type of colour and resemble real animals. The robots were continuously powered remotely from a laptop, leading to a platform that allowed the simultaneous control of multiple robots. Again, machine learning techniques were employed to construct a behavioural model from the real fish, which was then used to govern the behaviour of the robot fish. Effect of animal robots In both honeybee and zebrafish societies, scientists successfully established a closed loop between the animals and the robots. Animals could sense stimuli emitted by the robots and react to them, while the robots could also sense the animals around them and react to them. This facilitated self-organised swarm behaviour of animals and robots on both sides. Apart from learning the social language of swarming or flocking animals, the ASSISI_bf robots were designed to influence the collective behaviours of such groups. Long-term, this new technology is expected to help humans interfere with animal societies to better manage the environment. As Dr Schmickl explains, “the idea is that human operators can set goals for these communities, leading to applications in sustainable agriculture and livestock management.″ For example, understanding and influencing honeybee behaviour, from inside the hive for the first time, will help to develop new methods to protect this species. At the same time, it will help scientists explore and adapt the features that make honeybee societies so robust and efficient in their hybrid systems. Considering that animal swarms are encountered in our food chain but also in pests, the capacity to influence swarm behaviour has obvious benefits for food security and human health. Importantly, this information can potentially be employed to understand human societies.
Keywords
ASSISI_bf, robot, fish, swarm, honeybee, biomimetic, machine learning