Imagine an autonomous drone flying through complicated real-world terrain such as a dense wood, or the ruins of the Fukishima nuclear power plant. As drone technology currently stands the drone is likely to collide with a tree trunk or low hanging wire and crash to the ground rendering it useless without human intervention.
Before drones can be deemed safe to operate in densely populated areas or become truly useful in fields such as disaster relief they must be able to be able to be intelligent enough to autonomously avoid collisions, or even, as Dr Dario Floreano, director of the Swiss National Center of Competence in Robotics, has found, to be robust enough to be 'collision-friendly'.
Inspiration from nature
Floreano's research into the difficulties and challenges of creating drones in built environments has led him to take inspiration from the mechanisms that flying animals and insects have evolved to traverse safely and efficiently through complicated terrain.
To achieve similar agile flight drones must be lightweight, safe, able to perceive the environment using sensors to take suitable controls and actions, have excellent reaction times and be able to co-ordinate with other drones - 'swarm intelligence'.
Floreano set his sights on the compound eye found in insect vision for perception, as flying insects such as the common housefly use their specialised vision for several functions: altitude stabilisation, collision avoidance, altitude regulation, landing and chasing and homing. Compared with human eyes, convex compound eyes possess a wide-angle view. They can detect fast movement and, in some cases, the polarisation of light.
Floreano says: “We know that insects exploit optic flow as different contexts and features slide over the vision. The nervous system of insects has just a few neurons sitting a few synapses away from hundreds of optical photoreceptors. They put together motion information from different visual directions and when combined and analysed this is translated into motor movement.”
In an attempt to replicate this complex biological system for drones his research team built a simple mechanism that could gather real-time information using two cameras mounted on a 10g indoor micro-flyer, one looking at 'optic flow' on both the left and right side and another monitoring the ground. Gyroscopes were utilised for optic flow and an anemometer was used to measure the propellor rotations per minute for speed regulation.
Two sets of 'complex digital neurons' could analyse data, with one neuron used to perform evasive actions, and another neuron to analyse distance from ground and change in height. Simple controls, such as rudders and elevators were used like a traditional plane. Floreano says: “Put all this together and in a simple environment it can avoid the walls and ground with only two simple neural networks, and it took only few lines of code.”
Floreano's team quickly moved on to five cameras attached to a drone to measure optical flow in a specific directions, connected to the two neurons. The drone could easily avoid obstacles and structures by pitching up and down. The technology is already being used in commercial drones, such as senseFly, to maintain a safe distance from the ground and for visual-based landing.
Real world application
A handful of cameras and a few lines of code cannot replicate the complex compound eye found in flies. The large amounts of computational power needed, associated costs and weight caused by bulky cameras are also impractical.
To fly in cluttered environments Floreano and his research team had to drastically step up their game.
They turned to organic flexible electronics, so the electronics and photoreceptors placed on the opposite side could be bent to mimic the concave design of the compound eye. This meant it could grab more than one image at at time, extracting optic flow from it's surroundings just like a house fly, offering a very larger field view and no distortion, although the images do lose some resolution as a result.
Floreano says: “We put the flexible stacks together and cut down the compound all the way to flexible skin and bent it to create the eye less than one gram in width and 0.85mm deep.”
He compares the design to the dragonfly eye. It is still fairly large, cylindrical in shape and only bends in one direction. Floreano adds: “However, trilobites are the first example of the first compound eye, which is exactly the same shape and size as ours. Give us another 250 million years and we'll have our fly's eye.”
Collision-avoidance to collision-free
In the future Floreano predicts they will be able to have complex compound eyes in any configuration and density. Engineers will be able add flexible strips of compound eyes on any structure, from drones to objects such as hats, which could be used to aid the visually impaired using feedback vibrations.
However, Floreano realises that no matter how good the eye, sooner or later a drone will be unable to detect a sudden obstacle and will crash and fall. He says: “What happens when insects fly in confined spaces? They collide all the time. They then use their wings and legs to upright themselves and maintain balance or use legs on the ground. We also need to design drones not only for collision avoidance but collision resilience too.”
A gimball drone
They developed a drone for emergency situations and disaster relief built with a surrounding flexible cage to absorb collision. The drone also features legs to enable it to upright itself when on the ground.
The design was more basic than the collision-avoidance drones, with the rotor-copters having just four photodiodes to detect and fly towards light. However, given the basic task of flying down a hallway to reach a light source, the drones took a long time and a lot of energy to reach its goal, often falling on the ground when hitting obstacles.
To overcome this the team designed a drone with two cages, including an inner cage that houses and protects the propellers, a coaxial motor, two control surfaces, the battery, an IMU and control electronics. In case of collision, the GimBall’s spherical protective frame prevents obstacles from touching the inner frame and can passively rotate thanks to a gimbal system.
The system allows the GimBall to fly in close proximity to humans. The simple solution is also lightweight – ideal for long distance inspection drones needed for disaster relief missions or industrial inspections.
The research team tested the drone guided by sensing a magnetic route through a forest and it travelled for several hundred maters withstanding multiple collisions with trees. Using funds from the Swiss National Centre of Robots, Floreano has spun out a company called Flyability, which will shortly begin shipping its first commercial collision-friendly GimBall drones.
Next, Floreano says he wants to develop fixed wing drones for disaster relief that could transform into a walking structure and then take off again, much like a vampire bat, which changes the morphology of its body to use its wings as legs.