DE L'ABEILLE AU ROBOT : LA RÉGULATION DU FLUX OPTIQUE. Contrôle conjoint de vitesse et d'évitements d'obstacles latéraux pour véhicules totalement actionnés.

• Serres Julien

• Optic flow
• Honeybee
• Micro-air vehicles MAV
• Unmanned aerial vehicles UAV
• Insect flight
• Bionics
• Biorobotics
• Biomimetics

We developed an autopilot, called LORA (Lateral Optic flow Regulation Autopilot), which is inspired by motion vision in flying insects. It incorporates two interdependent optic flow regulators, each of which controls one translational degree of freedom: a bilateral optic flow regulator controls the robot's speed, while a unilateral optic flow regulator makes the robot avoid lateral obstacles. Simulation experiments show that a fully actuated vehicle, equipped solely with the LORA autopilot, is able to cross straight, tapered or even non stationary corridors. The robot needs only a pair of lateral eyes that measure the right and left optic flows, and requires no speed or range sensors. This autopilot is meant to equip a miniature seeing hovercraft (0.8 kg) equipped with two elementary (2-pixel) eyes and rendered fully actuated by two lateral thrusters. We identified experimentally all the robot's dynamical parameters and incorporated them into the simulation. This work is a first step toward a deft, lightweight and power-lean guidance system for micro-air vehicles (MAVs). Designing the autopilot involved three steps: LORA I, LORA II and LORA III that were progressively informed by the results of behavioural experiments carried out on bees trained to enter various corridors en route to an artificial nectar source. High resolution video recording and statistical analyses of the bees' trajectories in various environments allowed the underlying visuomotor control system to be deciphered. We challenge the "optic flow balance" hypothesis that was put forward 20 years ago to explain the bee's "centering reaction" in a corridor. Even though this behaviour has inspired the design of many robots eversince, it now appears as a degenerate case of the insect's ''wall following behaviour''. Our work involved constant plying between Biology to Robotics and provides spin-offs in both fields. It allows one to better understand how a 100mg insect can navigate without using any SONAR, RADAR, LIDAR or GPS, while offering autonomous vehicles an opportunity to behave the same way, without any needs to measure speed and range.