Caltech Building Agile Humanoid Robot by Combining Legs With Thrusters

Leonardo augments humanoid legs with thrusters to help it run and jump

Leonardo augments humanoid legs with thrusters to help it run and jump
Photo: Caltech
Advertisement

For better or worse, robots with humanoid features are often compared to humans—we want to know if they’re anywhere close to doing the same kinds of things that we do, and with a few exceptions, the answer is “probably not.” Humanoid robots are difficult to build and program, but we keep doing it because it makes some amount of sense to have robots that look and function like we do operating in the same environments that we operate in. However, one of the great things about robots is that they don’t have to be constrained by the same boring humanoid-ness that we are, and we can do all kinds of things to them to make them more capable than we’ll ever be.

Over the past year, we’ve seen several different projects that are enhancing the capabilities of humanoid robots by augmenting them with integrated thruster systems. There’s Jet-HR1, which uses foot-mounted ducted fans to cross large gaps, and of course IIT’s jet-powered flying iCub project

At Caltech’s Center for Autonomous Systems and Technologies (CAST), researchers are in the process of developing a new robot called the LEg ON Aerial Robotic DrOne, or Leonardo, which in addition to having one of the most creative acronyms we’ve seen in a while, combined a bipedal robot with a dronelike thruster system for balancing, agility, and the ability to leap tall buildings in a single bound. 

Leo is mostly made of carbon fiber, and is about 0.75 meter tall. It weighs just 2.75 kilograms, which is light enough that the thrusters mounted on either side of its torso can lift the entire robot off the ground. Leo isn’t really designed to be a flying robot with legs, though—those thrusters are primarily used to enhance what Leo’s legs will be able to do. What you’re seeing in the video, for example, is the thrusters being used to actively balance Leo’s upper body, so that Leo’s legs don’t have to do all the work. The propellers are synchronized with the leg joints to move the torso up and down under closed-loop control. Leo can balance on its own using just its legs, but using the thrusters in tandem could help it to make more agile motions, especially over rough terrain or in push recovery. It may help to think of the thrusters as an antigravity system that is there mainly to counteract the weight of the robot.

Since the thrusters can, if necessary, take the robot’s entire weight, it’s possible that Leo will be much better at not falling over, since it could fly for a brief period to reorient itself. Longer term, the idea is that Leo will be doing a lot of jumping, with the thrusters significantly augmenting both height and distance, with a multimodal nature that will increase versatility, reliability, and endurance.

The concept from Leo came primarily from Caltech postdoc Alireza Ramezani, who is now an assistant professor at Northeastern. Ramezani will continue to advise students at Caltech as the Leonardo project continues at CAST, but at the moment, Caltech professors Soon-Jo Chung and Mory Gharib are overseeing things out in California. We spoke with them last week for more details.

IEEE Spectrum: Where did the idea for a robot like this come from?

Mory Gharib: For many applications that we’re thinking about for the future, like a flying ambulance project that we have or missions to Mars, there is a huge need for I would say a third party—a robotic partner that can, in very extreme situations, conduct scouting or help people in ways that that either drones or bipedal robots can’t do. That was the whole idea—we need to have a system that basically can defy gravity to go places where other robots cannot. And because this machine is not going to fly in the way that drones do, because it has most of the time its legs are on the ground, it can carry a much heavier battery and payload. 

Soon-Jo Chung: We want to lay the groundwork for a next generation agile robot that walk and run, and at the same time can jump very high to overcome obstacles and then make a safe landing. That will open up a lot of possibilities both on Earth and for space exploration. The technical challenge we’re going to address is synchronized control of fan arrays that are typically used for drone control, with bipedal motion.

It sounds like the thrusters will be an integral part of the balancing and movement system, as opposed to just something that kicks in if the robot needs to jump, right?

Soon-Jo Chung: Absolutely. Our idea is that it’s easier to have closed loop control with close coordination between fan array control and leg joint control to actually enable agile maneuvers that are not possible for typical bipedal robots. And we’re already doing that, as the video demonstrates.

Mory Gharib: You can imagine the challenge of such a design—because you’re working with thrusters, there will always be a delay, and that means the machine should be smart enough to predict when a situation will arise, and start the thrusters before. You don’t want them to be on all the time, because it makes interaction with humans more difficult.

If everything works perfectly, what kinds of capabilities will the robot have?

Soon-Jo Chung: Walking on flat terrain, walking, running, and jumping to overcome small obstacles by using the lift generated by the propellers. And it should be able to in a very soft and stable fashion land after it jumps or flies. The ultimate form of demonstration for us will be to build two of these Leonardo robots and then have them play tennis or badminton.

Mory Gharib: Consider the idea of the Mars helicopter that is being developed. If you look at what it can deliver, it’s very limited. It has to fly in very thin air, and it can’t go too far away because of power limitations. A system like Leo can take advantage of being able to carry more batteries, and can also jump and scout areas that a single helicopter drone cannot. That I think has become a source of discussion for us—is this a better way of doing it? We think it could solve many problems for a mission like this. And there’s also the cost—this is going to be much less expensive compared to heavily machined robots. It can be practically produced by additive manufacturing because of the lightness of the concept.

What are some potential applications for a robot like this?

Mory Gharib: We have a project, we call it Guardians. These are basically dormant autonomous robot systems that are watching the environment, and the whole idea behind them is they are gathering information to be able to do a limited interaction to remedy a situation while for example first responders arrive.

Think of the recent fire in California—if you had systems that could easily walk and fly in challenging environments, you could get to people that a human cannot, or drones cannot, or a slow walking robot cannot. This system can be very adaptive to environments, maybe making sure that a person gets the information or tools that they need to get out of harm’s way. These are the kind of applications where, if you look at robot manufacturers or drone manufacturers, they avoid it, because they know it’s difficult and not going to happen. We’re not saying that Leo will be able to do that, but this approach that’s flexible where a robot can decide when to use what mode of operation is the kind of thing that we’re going to test with Leo.


The first challenging environment that Leo is likely to encounter, we’re told, could be a hopscotch court, which we may see the robot conquering within the next few months. And beyond that, there are also plans to add tools to the robot, perhaps even arms, which it will definitely need if it’s going to play a successful game of tennis.

[ CAST ]

Robotics News

Biweekly newsletter on advances and news in robotics, automation, control systems, interviews with leading roboticists, and more.

About the Automaton blog

IEEE Spectrum’s award-winning robotics blog, featuring news, articles, and videos on robots, humanoids, automation, artificial intelligence, and more.