Due to their size and locomotion ability, cockroaches are favorable as hybrid robot platforms in search and rescue (SAR) missions. However, cockroaches most likely approach the corner area and stay for an uncertain time. This natural behavior will hinder the utilization of cyborg cockroaches in SAR missions under rubble, unstructured, and unknown areas. Therefore, we proposed onboard automatic obstacle avoidance and human detection that can run on the wireless backpack stimulator without an external motion capture system. A low-power and small-size Time of Flight (ToF) sensor was selected as a distance measurement sensor, while a low-resolution thermopile array sensor was applied for human presence detection. The implemented feedback control based on IMU and ToF sensors has successfully navigated the cyborg cockroach to avoid obstacles and escape from the sharp corners in the laboratory unstructured area without stopping or being trapped. It could also recognize the human presence when the human was in front of it in real-time. Due to its performance, the random forest classifier was implemented as an embedded human detection system. It could achieve the highest accuracy at a distance of around 25 cm (92.5%) and the lowest accuracy at about 100 cm (70%).
Keywords: Cyborg insect; Embedded machine learning; Human detection; Obstacle avoidance.
© 2024 The Authors.