Ultrasonic sensors let drones navigate like bats
A drone navigates using ultrasonic sensors and echo evaluation. This is less energy-intensive than conventional methods.
(Image: WPI)
A team of engineers from Worcester Polytechnic Institute (WPI) has equipped a mini-drone with ultrasonic sensors and artificial intelligence (AI), enabling it to navigate autonomously with low-energy and computational effort in poor visibility conditions such as darkness, fog, or smoke without collisions. The drone could be used, for example, in search and rescue operations in confined buildings.
Bats inspired the WPI researchers to develop a simpler navigation system for drones that additionally works in difficult visibility conditions and is not solely based on energy-intensive, expensive lidar sensors, radar, and cameras, which also make the drones heavier.
Smaller bats, such as the bumblebee bat, weigh less than 2 g. They orient themselves in dark, humid, and dusty caves by emitting short chirps and receiving their faint echo with a few neurons, says Nitin J. Sanket, a professor at WPI's Robotics Institute. Sanket and his colleagues' navigation system is based on this.
In the quadcopter drone, which measures approximately 16 cm diagonally and weighs around 460 g, the scientists have installed two tiny ultrasonic sensors of the type TDK InvenSense ICU30201, as described in the study “Milliwatt ultrasound for navigation in visually degraded environments on palm-sized aerial robots,” published in Science Robotics. The two sensors are synchronized with a Teensy 4.1 microcontroller.
Empfohlener redaktioneller Inhalt
Mit Ihrer Zustimmung wird hier ein externes YouTube-Video (Google Ireland Limited) geladen.
Ich bin damit einverstanden, dass mir externe Inhalte angezeigt werden. Damit können personenbezogene Daten an Drittplattformen (Google Ireland Limited) übermittelt werden. Mehr dazu in unserer Datenschutzerklärung.
Recognizing and evaluating ultrasound echoes
However, to reliably use the acoustic sensors for navigation, the researchers had to filter the echo signals of the ultrasonic sensors from the drone's propeller noise. First, they equipped the drone with acoustic shielding. To then evaluate the faint ultrasound echo signal pattern, the researchers trained an AI using deep learning, which interprets the incoming echo, similar to how bats do, to detect obstacles. For signal emission and data evaluation for navigation, the researchers used a Google Coral Mini developer board with Mendel GNU/Linux 5 as the operating system; the autonomy software runs under Robot Operation System 2 Humble.
The software evaluates the data and calculates the flight path with the necessary control signals via navigation algorithms. These are then sent to the drone's flight controller via a MAVLink protocol. This is done in real-time with low energy consumption. This allows the 4S lithium-ion battery with 850 mA to be small and the drone to be kept light overall. The power of the small battery used is sufficient for a flight time of about five minutes.
The researchers conducted various tests with the drone in the lab and in the field. In the lab, the drone had to avoid black obstacles in the dark and completely autonomously navigate an obstacle course in fog and simulated snowfall. In field tests, it flew through a forest with thin trees with small branches.
The scientists at WPI conducted a total of 180 tests. The drone achieved a success rate of 72 percent. At lower speeds of 1 m/s, the success rate was higher (100 percent) than at higher speeds (81.61 percent at 1.5 m/s and 72.73 percent at 2 m/s). Thin metal rods and particularly small branches proved to be difficult-to-detect obstacles. Both obstacles reflect ultrasound signals only very weakly, leading to lower detection performance.
Videos by heise
In addition to more reliable detection performance, the battery life of the bat drone now needs to be further increased, because in real search and rescue operations, a few more seconds of flight time could decide between life and death for a survivor, says Sanket.
(olb)
