Dissertation Defense

Bio-Inspired Optic Flow Sensors for Artificial Compound Eyes

Seo Jun Park

Compound eyes in flying insects have been studied to reveal the mysterious cues of vision-based flying mechanism inside the smallest flying creature in nature. Especially, researchers in the robotic area have made efforts to transfer the findings into their unmanned aerial vehicles in less than a palm size, micro-air-vehicles (MAVs). The miniaturized artificial compound eye is one of the key components in this system to provide visual information for navigation. Multi-directional sensing and motion estimation capabilities can give wide field-of-view (FoV) optic flows up to 360Â °~ of solid angle. By deciphering the wide FoV optic flows, relevant information on the self-status of flight is parsed and utilized for flight commands generation. In the last few years, several artificial compound eyes have been demonstrated by forming a hemispherical lens configuration to realize an independent optical path to each photoreceptor. However, they require complicated fabrication processes, limiting their applications due to poor yield and high cost. In this work, we take a simple and practical approach. We realize the wide-field optic flow sensing in a pseudo-hemispherical configuration realized by mounting a number of 2D array optic flow sensors on a flexible PCB module. The flexible PCBs can be bent to form a compound eye shape by origami packaging. From this scheme, the multiple 2D optic flow sensors can provide a modular, expandable configuration to meet low power constraints. The 2D optic flow sensors satisfy the low power constraint by employing a novel bio-inspired algorithm. We have modified the conventional elementary motion detector (EMD), which is known to be a basic operational unit in the insect's visual pathways. We have implemented a bio-inspired time-stamp-based algorithm in mixed-mode circuits for robust operation. By optimal partitioning of analog to digital signal domains, we can realize the algorithm mostly in digital domain in a column-parallel circuits. Only the feature extraction algorithm is incorporated inside a pixel in analog circuits. In addition, the sensors integrate digital peripheral circuits to provide modular expandability. The on-chip data compressor can reduce the data rate by a factor of 8, so that it can connect total 25 optic flow sensors in 4-wired Serial Peripheral Interface (SPI) bus. The packaged compound eye can transmit full-resolution optic flow data through the single 3MB/sec SPI bus. The fabricated 2D optic flow prototype sensor has achieved the power consumption of 243.3pJ/pixel and the maximum detectable optic flow of 1.96rad/sec at 120fps and
60Â °~ FoV.

Sponsored by

Euisik Yoon