A team of European researchers has developed a robot with an ?artificial mini-brain,? based on honeybees, that allows it to learn how to react to environmental stimuli.
?NeuroRover? is a small, wheeled robot 25cm (9.8 inches) long and 10cm (3.9 in) wide fitted with two cameras to perceive colors and a light sensor to detect rewards.
Once the robot learns to make specific associations – for example, associating red objects with reward, just as the honeybee associates certain flower colors with nectar – the motor network’s activity patterns change and the little robot trundles autonomously towards its goal.
The end result is a robot that links specific environmental stimuli with behavioral rules, but it’s the bio-inspired structure of the artificial intelligence (AI) that has the most potential for widespread application in robotic systems.
Unlike ?conventional? intelligent systems, which are ?remotely inspired by brain computation,? the researchers followed a ?radically different paradigm,? Martin Paul Nawrot, professor of neuroscience at Freie Universitat Berlin, told Robotics Business Review.
?We simulated artificial mini-brains with networks of actual neurons – that is, spiking neurons that integrate and generate electrical signals just as real biological neurons – that are connected via actual synapses in an anatomical fashion,? says Nawrot.
The team has no specific short-term commercial applications in mind, says Nawrot, but neural processing systems could bring inherently parallel, fast, and energy-efficient AI to a wide range of autonomous robots.
?Just like their animal idols, adaptive neural systems could learn the structure of their environments. On a more general level, brain-like processing is of interest for many intelligent systems for the same advantages: speed, energy-efficiency, and adaptation,? says Nawrot, who adds that insects can easily perform a number of complex behavioral tasks that are still beyond the capabilities of today?s AI systems.
NeuroCopter Taking Off
The team is also working on implementing the artificial brain on a quadrocopter (?NeuroCopter?), allowing it to explore and map unknown terrain.
The researchers used neural simulation software (IQR) to simulate a mini-brain in form of a spiking neural network that interacts in real-time with the robot’s sensors and actuators.
For future applications, however, the team hopes to replace the computer simulation with neuromorphic chips – silicon chips that embed networks of neurons and synapses.
The project is an interdisciplinary collaboration between the “Neuroinformatics” group led by Nawrot, the “Artificial Intelligence” group led by Raul Rojas, and Tim Landgraf’s Biorobotics lab at Freie Universitat Berlin. The work was published in an academic paper for the 6th International IEEE/EMBS Conference on Neural Engineering (NER).