Spanish Researchers Create a Neuromorphic Sensor That Sees Depth With a Single Lens

IKERLAN, a Spanish technology center, has developed a prototype neuromorphic machine vision system called BEGI that can simultaneously determine spatial depth and track motion using only a single lens and sensor. The device mimics the compound eye of an insect and is intended for so-called physical artificial intelligence, including robots, drones, and other autonomous machines that need to quickly perceive and interact with the surrounding world. Unlike lidars and depth cameras, the system operates passively, meaning it does not emit laser or infrared pulses. The prototype is based on a dynamic vision sensor, the Sony-Prophesee IMX636, over which a matrix of microlenses is placed. Each pixel of such a sensor transmits data not at regular intervals but only when it detects a change in brightness, forming a stream of individual events. The microlenses allow the direction of incoming light to be determined, and an algorithm for direct stereo reconstruction calculates object shapes, depth, and optical flow without a second camera. The IMX636 has a resolution of 1280 x 720 pixels, a latency of less than 100 microseconds at 1000 lux illumination, a dynamic range exceeding 120 dB, and a maximum power consumption of about 205 mW. The main limitation of the current sample is its resolution: to determine depth, the same point must be registered by multiple pixels at different angles. Therefore, the range of the practical version of BEGI is currently less than one meter, and its field of view is roughly 20 to 40 degrees. This limits the application of the sensor platform to near-range tasks, including manipulators on industrial lines, monitoring moving mechanisms, and the final docking stage of spacecraft, where the sensor must almost instantly detect small displacements under strict power and heat-dissipation constraints. Currently, the computational part of BEGI is implemented on a programmable FPGA matrix and produces filtered event frames, dynamics, optical flow maps, and depth maps. The next step is to develop a specialized ASIC chip and a large-format sensor with significantly more pixels. According to the developers' estimates, such a version could increase sensing range to approximately 16 meters and expand the field of view to 120 degrees, while maintaining microsecond-level response and a single-lens design. This would allow the technology to be widely used in mobile robots, navigation systems, drones, and autonomous vehicles. But will it happen? The developers hope they can make it through this path.