Two students of the Department of Aerospace Engineering, IIT Madras have designed an AI-run drone that can secure air space over critical civilian and military installations and can ward off rogue surveillance drones. The drone can also track down rogue drones visually, hack into their GPS navigation system, following which the target drone is forced to change its flight path or land safely. All of this can be controlled remotely over the internet.
Law enforcement agencies, security services and armed forces can now use this tech to secure critical air space from surveillance by rogue drones. A major advantage of this system is that it can be controlled over the internet and can navigate autonomously as compared to most existing drones that operate on ‘line of sight' meaning the operator must keep the drone within their sight. Using the internet to control the drones also allows for deploying a swarm of drones that can intelligently detect and track people, drones, vehicles and other objects.
Vasu Gupta, a final year BTech student, Department of Aerospace Engineering, IIT Madras, and Rishabh Vashistha, a Project Associate working in the RAFT Lab of the same department have developed the prototype. The researchers designed a visual-based tracking system using Deep Neural Networks (Artificial Intelligence) to secure airspaces and land stretches efficiently by employing a swarm of drones. The motion detection algorithms are powered by AI and can detect motion even in dark conditions without the need of an IR (infrared) camera, explained the students.
Further explaining the functioning of these drones, Vasu said, “The drone works by employing a software-defined radio and broadcasting spoofed GPS signal by making use of the ephemeris data of GNSS constellations. The target drone’s GPS sensor locks onto our fake radio station transmitting at a much higher power than the available satellite’s transmission power. Following this, the drone generates fake GPS packets by mathematically modelling the time differences at the receiver’s end. Using four such time differences, the GPS sensor calculates its 3D position and calibrates the rogue drones’ time to our spoofed clock. This way, we alter the latitude, longitude, altitude and time of the rogue drones.”
The team used an advancement of Kernelized Correlation Filters for tracking objects once they are detected and locked onto. Such tracking features work on visual sensors like cameras and CMOS without using radars and sonars, the latter of which generally do not provide much informative data. The duo was mentored by Dr Ranjith Mohan while developing this project. Dr Mohan said, “Our current prototype is equipped to detect and track objects visually, precisely land and fly over the internet. Our next step will be to conduct exhaustive tests on the system and ensure its reliability for catering to a wide range of demanding missions that pose challenge to our law enforcement and defence agencies. The programmable nature of our aerial vehicles also opens up the possibility of swarming multiple vehicles to act as a team and accomplish a common mission.”
Rishabh added that algorithmically altering the 3D position allows the drone to move the target drone locally. "Moreover, when a large variance is given in the spoofed GPS position, a failsafe (if any) is invoked on the target side which results in a safe landing of the target drone,” he said. “We have tested this electronic countermeasure of ours against nearly all the civilian GPS receivers used by the UAV industry such as ublox and DJI inhouse GNSS and we have been able to take down the drones almost instantaneously (within 4-5 seconds),” added Rishabh.
