‘Ornithopers have been around for many years and robots based on the concept are not new. But, according to a UMD news release, researchers there have made some new breakthroughs on this old idea by finally designing a robot bird that independently controls the flapping of each wing. This means the robot can fly much like a real bird, adapting to wind speed changes and performing aerobatic maneuvers. From Professors S. K. Gupta’s blog post about the project:
“Real birds are able to precisely control each wing during flight which enables them to do all sorts of aerobatic maneuvers. This has been a very difficult feat to achieve in bird-inspired robots. In fact, prior efforts (including our own) utilized only simple wing motions where both wings are driven by a single motor. So motions of two wings are coupled. Minor adjustments can be made in wing motions by using small secondary actuators. But two wings cannot move completely independently. In the past, any major change in the wing motion had to be accomplished by doing a hardware change on the ground. Clearly this limited how close a robotic bird came to the real bird in terms of the flight characteristics.”‘
More information is available at Dr. Gupta’s blog.
Dr. Gupta mentions that he and his team adopted the following innovations:
‘The new design uses two actuators that can be synchronized electronically to achieve motion coordination between the two wings. The use of two actuators required a bigger battery and an on-board micro controller. All of this makes our robotic bird overweight. So how do we get Robo Raven to “diet” and lose weight? We used advanced manufacturing processes such as 3D printing and laser cutting to create lightweight polymer parts to reduce the weight. However, this alone was not sufficient. We needed three other tricks to get Robo Raven to fly. First, we programmed wing motion profiles that ensured that wings maintain the optimal velocity during the flap cycle to achieve the right balance between the lift and the thrust. Second, we developed a method to measure aerodynamic forces generated during the flapping cycle. This enabled us to quickly evaluate many different wing designs to select the best one. Finally, we had to perform system level optimization to make sure that all components worked well as an integrated system.‘
The entire project was a very impressive feat of engineering. Indeed, this was an outstanding instance of true innovation. One question that comes to mind is if evolutionary algorithms could have been used both in the design process and for letting the mechanical bird learn to fly. For example, Gupta mentions the ability to evaluate wing designs. While he does not provide details, I assume that he was able to do this via computer simulation first then verify results with physical experiments (wind tunnel, etc.). If so, then this is the type of problem that is well suited to evolutionary algorithms.
Teaching the mechanical bird to fly via evolutionary algorithms would have been interesting. This has been done to teach robots how to walk, but flight would probably be more difficult. Also, this would require that the robot’s brain be based on a separate computer with wireless signals being sent to the little brain on board. Gupta mentions this in a post about cloud computing for robotics, “Cloud Robotics: Are We Ready to Put the Robot Brain in the Cloud?“. There is no need to use the cloud per se. It is sufficient and in fact more convenient, to use a dedicated computer or network. The only advantage to using the cloud is that it is cheaper, but this reduction in costs comes with the inconvenience of not having exclusive control and being exposed to the whims and vagaries of the cloud owner. As computing continues to decline in costs in real terms, evolutionary algorithms become more attractive as their inherent slowness becomes more manageable.