Bio-inspired source seeking and obstacle avoidance on a palm-sized drone

Abstract

The small form factor of palm sized unmanned aerial vehicles (UAVs) combined with their ability to freely maneuver in 3D space with holonomic trajectories and carry custom sensors makes them an ideal platform for autonomous source seeking in challenging environments. Equipped with the appropriate sensor, a small UAV could autonomously navigate towards light or heat sources such as forest fires or locate a radio-frequency (RF) transmitter attached to anything from a package in a warehouse to an animal tagged with a radio tracker. Leveraging small UAVs for this task however requires addressing their size weight, power, and computational constraints. While prior source seeking robots have used search strategies that require extensive training, such as reinforcement learning, we instead look to biology and employ a simple ‘run and tumble’ gradient following algorithm inspired by bacterial chemotaxis. The result is a computationally inexpensive approach requiring as little as 30 instructions/second, allowing this strategy to scale down to millimeter scale robots with small microcontrollers. Using insights from simulation, we report a success rate of 91% in real-time demonstrations of our UAV navigating towards a fire or light source while avoiding obstacles. Measurements from a small Bluetooth transmitter indicate it also produces a compatible gradient at ranges of 50–100 m. We conclude by discussing how this technique could scale down to sub-cm microrobots seeking RF power sources.