Tarantula hawk uav

It may sound like the stuff of science fiction, but it has been the goal of real-life military research for decades, arguably beginning in earnest in the 1970s with the CIA’s attempt to perfect a gas-driven robot dragonfly – the ‘insectothopter’ – to help its covert intelligence gathering operations. Although that project ultimately came to nothing, forty years on new developments in microelectromechanical systems (MEMS) and DARPA’s recently announced Fast Lightweight Autonomy (FLA) programme mean that robotic spy-flies are now closer than ever to becoming a reality. Nevertheless, there are still some challenges ahead.

"Scientists and engineers have a reasonable understanding of the aerodynamics of insect-like and insect-scale flight; however, we’re still significantly limited by the power, sensing and computing resources that can be put on such small aircraft," says Dr Larry Matthies, senior research scientist and supervisor of the Computer Vision Group at the Jet Propulsion Laboratory in Pasadena, US. "There are also limitations in the performance of actuators and drive-trains to move the wings efficiently."

On robotic wings

In December 2014, however, the US Army Research Laboratory (ARL) unveiled what seems to represent a significant breakthrough in actuator design – a pair of robotic insect wings just three centimetres long, made of lead zirconium titanate, which bend and flap when a voltage is applied. According to Dr Ron Polcawich, who leads the ARL team working on piezoelectric microelectromechanical systems (PiezoMEMS), tests have shown that not only do they flap like insect wings, they can also generate lift.

There are obvious similarities to Harvard University’s RoboBee which, with its tiny 3cm wingspan, is said to be the world’s smallest flying robotic insect, and is, as the ARL researchers readily acknowledged, further advanced along the road to independent flight – but their goal is a robot around an order of magnitude smaller. At that level, however, there is a whole series of issues to overcome.

The challenge of scale

ARL’s MEMS design and mechanical engineer Jeffrey S. Pulskamp explains that although many of the individual components have been demonstrated, much research is still required to enable the integration of all the sensor, control, processing and actuator technologies, and then to scale everything down to the necessary size.

Whole new groups of robotic agents are now evolving, inspired by the world of nature.

"For example, there are many challenges just within the context of mobility. Developing actuators that not only can mimic movement of insects but also address robustness to interacting with real world environments," he says.

Living insects produce the majority of their thrust during the downward movement of their wings – termed the power stroke – and the most agile flyers amongst them can flap those wings 120 times a second. Mimicking this calls for actuators that can provide a relatively powerful, large air displacement, are capable of high-frequency operation and ideally require low voltages to achieve it. It is a big jump, but Pulskamp says that innovative technologies could hold the key.

"The most significant developments in recent years have been in miniaturised actuation technologies, the fabrication processes to enable them, and the development of fast and reflexive bio-inspired control systems. In regards to our research, our advancements were made possible based on ARL’s expertise in PiezoMEMS technology," he says.

Cutting the tether

While the strides taken in solving the mechanics of micro-robot flight are undeniably impressive, there is more than this to producing a truly deployable spy-fly. One concern, at least for the moment, is the question of an onboard power source; even the relatively advanced RoboBee currently remains firmly tethered to the ground by a wire. Looking beyond that, there remains a more recalcitrant conundrum to solve: How do you provide the necessary sensing and computing capabilities to enable robot wings to be used autonomously, when the payload of insect-sized aerial vehicles will always be tiny?

DARPA hopes its FLA programme, launched with a call for proposals just before Christmas last year, will go some way towards answering that question. The programme aims to create a new class of algorithms that will allow small UAVs to navigate independently around rooms, corridors and stairways and through obstacle-filled environments, quickly and efficiently – at up to 20m/s – without the need for remote pilots, GPS data or information from external sensors.

Thinking like a bug

The goal is to equip the drones with the necessary sensing, perception, planning and control elements to allow UAVs to mimic the navigational and autonomous flying skills of birds and insects. If successful, it will be the first step towards creating a robot insect that not only flies like a bug, but thinks like one, too. That step is, however, a huge one.

"It may sound like the stuff of science fiction, but it has been the goal of real-life military research for decades."

The DARPA brief specifies a UAV small enough to fit through an open window – not something insect-sized, or indeed insect-shaped – and the difficulties of achieving even bug-like levels of situational awareness, never mind the cognitive abilities of a bird, in a drone that small are already well documented.

In the CVPR2014 workshop paper titled ‘Towards autonomous navigation of miniature UAV’ Matthies and his co-authors describe the problems involved in getting a mini-UAV, in this case a twelve-gram quadcopter, to orientate itself and then identify a route. "Fast and accurate pose estimation, that is independent from external sensors (e.g. GPS), is needed for control," they write. "Literature has shown that a viable solution for GPS independent pose estimation is to use visual and inertial sensors. However a major algorithmic challenge is to process sensor information at high rate to provide vehicle control and higher level tasks with real-time position information and vehicle states."

To put it simply, the brain that a bird-sized drone can currently carry struggles to crunch the numbers fast enough to come up with a bug-style performance in the air – but things are slowly changing. The work that Matthies and his colleagues have done on novel, low-resource algorithms is a clear signpost on the road to autonomous mini-UAVs and perhaps, ultimately, robot insects.

Maturing the technology

The big remaining question is: When will the physical systems required for flying, and the artificial brain-power needed to control it, be advanced enough to let the first generation of robo-bugs actually take to the skies?

Research is ongoing on all of those topics, Matthies says, adding: "but I don’t have a clear idea of how long it might take to mature; many years in any case."

Scientists at Georgia Tech were inspired by the animals and people who always land on their feet.

Pulskamp agrees, saying: "Our basic research in this area is intended, in part, to assess the feasibility of building such systems. This is an ongoing process. If it is possible to eventually build useful systems, I doubt it will be in less than 20 to 30 years."

Even if it takes twice as long as that to produce functional robotic insects, it would still be just the blink of an eye in terms of the time it took nature to develop the original models. According to the findings of an international study published in the 7 November’s issue of Science, although real insects appeared a lot earlier than was first believed – about 479 million years ago – it still took them a further 73 million years to develop their remarkable powers of flight.

The evolution of robo-insects is looking positively speedy by comparison.

Defence link