Released on
May 29, 2025
Read
4 min
Category
Industry Insight
Animal inspiration
Song’s feline-like eye is not his only design that has been inspired by the natural world. He and his colleagues were similarly influenced by the eyes of the cuttlefish, the pupils of which take on a ‘W’ shape under bright light. In an unevenly lit ocean environment, a W-shaped pupil reduces the amount of light entering the eye from the bright surface of the water.
Brain-like sensing
CurvACE’s innovations go beyond the shape and wide field of view: the design of the image sensors also diverges from convention. CurvACE focuses light onto an array of neuromorphic photodetectors, so named because their design is inspired by the neural architecture of the fly’s eye. The output of the photodetectors signals the intensity of light perceived, which the eye is programmed to process as optic flow. Each ommatidium is equipped with a neuromorphic adaptation circuit, which enables the photodetectors to accurately respond to a much broader range of light intensities than can the sensors of conventional cameras, and at a much higher frequency. This allows the system to function well in a weakly lit room or under a bright sky, just like an insect’s eye.
Bodily improvement
The big challenge with using neuromorphic cameras in robotic vision is aligning the sensing with the computing. Conventional vision algorithms, which take in image sequences, are not designed to work with event data. This means that the output of neuromorphic sensors might need to be converted into a form that the software can understand — defeating the purpose of using an unconventional sensor in the first place. “You will have to make frames out of events, and that will again slow you down,” says Sandamirskaya. “You lose all the advantages.” For this reason, she thinks that it is important to “couple this camera with the computer that also fits”. But having been around for only a few decades, there is much work still to be done to develop an ecosystem around these cameras. “Neuromorphic hardware in general is still quite experimental, and not always easy to work with due to interfacing limitations,” says Jesse Hagenaars, an artificial intelligence (AI) and robotics researcher at Delft University of Technology in the Netherlands.
