A couple of years ago, we wrote about a bipedal robot called Jet-HR1 under development at the Guangdong University of Technology. With little foot-mounted ducted fans, Jet-HR1 could step across very wide gaps by using the thrust created by the fans to futz with its center of gravity. That's cool and all, but let's take the logical (or not!) next step and see what happens when those ducted fans get cranked up as high as they'll go: flying humanoid robot! Sort of!
This is obviously just the first tentative little airborne hop, but by the end of the video, you can see that the stabilization works pretty well. I wouldn't call it completely controllable yet, but it's tangible progress.
Jet-HR2 has 10 degrees of freedom for ground locomotion, plus four ducted fans, two statically mounted on the robot's waist and two mounted inside the feet that can be actuated through ankle movements. Each fan can deliver 5 kg of thrust, for 20 kg total, enough to lift the 17 kg robot. The thrust to weight ratio here is not great, which is where the control challenge is; without a lot of spare oomph, you have to be very careful about how you allocate thrust. But the system that you see in the video is able to effectively suppress diving and spinning, leading to a stable (although not entirely under control) flying most-of-a-humanoid robot.
A word here on practical applications—there aren't a heck of a lot of good reasons to make a humanoid robot in the first place. So why, then, is a flying humanoid robot actually useful? Or does it get a pass because, I mean, c'mon, a flying humanoid robot, right? Here's what the paper says:
Recently, various disaster-response humanoid robots have been invented with unique control theories and other mechanisms to overcome uneven terrain. Traditionally, humanoid robots have overcome these obstacles by stepping and climbing yet these strategies lack efficiency, especially for dangerous environments like insurmountable obstacles and geological faults. For urgent tasks in complex real scenarios, humanoid robots are expected to have dynamic aerial skills, such as high or long jumps, short distance flights, and hovering that exceed the body length several times.
The performance of humanoid robots is still not up to the human level, especially with an increase in mass. On the other hand, even at the human level, robots may appear helpless on loose, collapse prone, or cliff-like terrain. This seems to be a limitation of using purely joint actuators to generate force. In this study, a novel humanoid robot that can fly using a ducted fan propulsion system was developed to explore its potential value for search and rescue in complex environments.
Frequent readers of this site may have seen this one coming: robots for disaster relief and search and rescue tend to be the catch-all justifications for weird mobility concepts without immediately obvious applications. But on the other hand, this is actually one of the reasons why making a humanoid might be a good idea, because having robots that can go where humans go can be very helpful. That is, if you can get them to work, which you probably can't, because practical humanoid robots are super duper hard. What's not hard is imagining how a humanoid robot that can fly could be even more useful. Again, there's that whole getting it to actually work thing, but it's not completely crazy to do some of the foundational research to see what might eventually be possible.
Design of a Flying Humanoid Robot Based on Thrust Vector Control, by Yuhang Li, Yuhao Zhou, Junbin Huang, Zijun Wang, Shunjie Zhu, Kairong Wu, Li Zheng, Jiajin Luo, Rui Cao, Yun Zhang, and Zhifeng Huang, from Guangdong University of Technology, is available on arXiv.