Caltech’s LEO Flying Biped Can Skateboard and Slackline



Again in February of 2019, we wrote a few type of humanoid robotic factor (?) below growth at Caltech, known as Leonardo. LEO combines light-weight bipedal legs with torso-mounted thrusters highly effective sufficient to elevate all the robotic off the bottom, which might handily handle on-ground dynamic balancing whereas additionally enabling some slick aerial maneuvers.

In a paper printed immediately in Science Robotics, the Caltech researchers get us caught up on what they have been doing with LEO for the previous a number of years, and it might now skateboard, slackline, and make dainty airborne hops with exceptionally elegant landings.

These heels! Looks like an actual sponsorship alternative, proper?
The model of LEO you see right here is considerably completely different from the model we first met two years in the past. Most significantly, whereas “Leonardo” used to face for “LEg ON Aerial Robotic DrOne,” it now stands for “LEgs ONboARD drOne,” which will be the first even reasonably profitable re-backronym I’ve ever seen. In any other case, the robotic has been utterly redesigned, with the model you see right here sharing zero elements in {hardware} or software program with the 2019 model. We’re informed that the previous robotic, and I am quoting from the researchers right here, “sadly by no means labored,” within the sense that it was far more restricted than the brand new one—the previous design had promise, but it surely could not actually stroll and the thrusters had been solely helpful for leaping augmentation versus sustained flight.

To allow the brand new LEO to fly, it now has a lot lighter weight legs pushed by light-weight servo motors. The thrusters have been modified from two coaxial propellers to 4 tilted propellers, enabling angle management in all instructions. And every part is now onboard, together with computer systems, batteries, and a brand new software program stack. I notably love how LEO lands right into a strolling gait so gently and elegantly. Professor Quickly-Jo Chung from Caltech’s Aerospace Robotics and Management Lab explains how they did it:

Creatures which have greater than two locomotion modes should study and grasp easy methods to correctly swap between them. Birds, for example, bear a fancy but intriguing conduct on the transitional interface of their two locomotion modes of flying and strolling. Equally, the Leonardo robotic makes use of synchronized management of distributed propeller-based thrusters and leg joints to comprehend clean transitions between its flying and strolling modes. Specifically, the LEO robotic follows a clean flying trajectory as much as the touchdown level previous to touchdown. The ahead touchdown velocity is then matched to the chosen strolling pace, and the strolling section is triggered when one foot touches the bottom. After the landing, the robotic continues to stroll by monitoring its strolling trajectory. A state machine is run on-board LEO to permit for these clean transitions, that are detected utilizing contact sensors embedded within the foot.

It is very cool how Leo neatly solves among the most troublesome issues with bipedal robotics, together with dynamic balancing and traversing massive adjustments in peak. And Leo may do issues that no biped (or human) can do, like really fly brief distances. As a multimodal hybrid of a bipedal robotic and a drone, although, it is essential to notice that Leo’s design consists of some vital compromises as properly. The robotic needs to be very light-weight with a view to fly in any respect, which limits how efficient it may be as a biped with out utilizing its thrusters for help. And since a lot of its balancing requires energetic enter from the thrusters, it’s totally inefficient relative to each drones and different bipedal robots.

When strolling on the bottom, LEO (which weighs 2.5kg and is 75cm tall) sucks down 544 watts, of which 445 watts go to the propellers and 99 watts are utilized by the electronics and legs. When flying, LEO’s energy consumption virtually doubles, but it surely’s clearly a lot quicker—the robotic has a value of transport (a measure of effectivity of self-movement) of 108 when strolling at a pace of 20 cm/s, dropping to fifteen.5 when flying at 3 m/s. Evaluate this to the price of transport for a median human, which is properly below 1, or a typical quadrupedal robotic, which is within the low single digits. Probably the most environment friendly humanoid we have ever seen, SRI’s DURUS, has a value of transport of about 1, whereas the rumor is that the price of transport for a robotic like Atlas is nearer to twenty.
Long run, this low effectivity might be an issue for LEO, since its battery life is nice for less than about 100 seconds of flight or 3.5 minutes of strolling. However, explains Quickly-Jo Chung, effectivity hasn’t but been a precedence, and there is extra that may doubtlessly be accomplished to enhance LEO’s efficiency, though all the time with some compromises:
The intense balancing potential of LEO comes at the price of constantly working propellers, which ends up in increased vitality consumption than leg-based floor robots. Nonetheless, this stabilization with propellers allowed the usage of low-power leg servo motors and light-weight legs with flexibility, which was a design alternative to reduce the general weight of LEO to enhance its flying efficiency.There are attainable methods to enhance the vitality effectivity by making completely different design tradeoffs. As an illustration, LEO might stroll with the lowered help from the propellers by adopting finite ft for higher stability or increased energy [leg] motors with torque management for joint actuation that might enable for quick and correct sufficient foot place monitoring to stabilize the strolling gait. In such a case, propellers could must activate solely when the legs fail to take care of stability on the bottom with out having to run constantly. These options would trigger a weight enhance and result in the next vitality consumption throughout flight maneuvers, however they’d decrease vitality consumption throughout strolling. Within the case of LEO, we aimed to realize balanced aerial and floor locomotion capabilities, and we opted for light-weight legs. Reaching environment friendly strolling with light-weight legs just like LEO’s continues to be an open problem within the subject of bipedal robots, and it stays to be investigated in future work.

A rendering of a future model of LEO with fancy yellow skins
At this level in its growth, the Caltech researchers have been focusing totally on LEO’s mobility programs, however they hope to get LEO doing helpful stuff out on this planet, and that just about actually means giving the robotic autonomy and manipulation capabilities. In the intervening time, LEO is not notably autonomous, within the sense that it follows predefined paths and does not determine by itself whether or not it needs to be utilizing strolling or flying to traverse a given impediment. However the researchers are already engaged on methods through which LEO could make these choices autonomously by means of imaginative and prescient and machine studying.

As for manipulation, Chung tells us that “a brand new model of LEO might be appended with light-weight manipulators which have related linkage design to its legs and servo motors to increase the vary of duties it might carry out,” with the objective of “enabling a variety of robotic missions which might be arduous to perform by the only use of floor or aerial robots.”
Maybe probably the most well-suited purposes for LEO can be those that contain bodily interactions with buildings at a excessive altitude, that are normally harmful for human staff and will use robotic staff. As an illustration, excessive voltage line inspection or monitoring of tall bridges might be good purposes for LEO, and LEO has an onboard digital camera that can be utilized for such functions. In such purposes, standard biped robots have difficulties with reaching the positioning, and normal multi-rotor drones have a difficulty with stabilization in excessive disturbance environments. LEO makes use of the bottom contact to its benefit and, in comparison with a regular multi-rotor, is extra immune to exterior disturbances reminiscent of wind. This may enhance the protection of the robotic operation in an outside surroundings the place LEO can keep contact with a inflexible floor.
It is also tempting to take a look at LEO’s potential to roughly simply bypass so most of the challenges in bipedal robotics and take into consideration methods through which it might be helpful in locations the place bipedal robots are inclined to battle. Nevertheless it’s essential to keep in mind that due to the compromises inherent in its multimodal design, LEO will seemingly be greatest fitted to very particular duties that may most straight leverage what it is notably good at. Excessive voltage line and bridge inspection is an efficient begin, and you may simply think about different inspection duties that require stability mixed with vertical agility. Hopefully, enhancements in effectivity and autonomy will make this attainable, though I am nonetheless holding out for what Caltech’s Chung initially promised: “the last word type of demonstration for us can be to construct two of those Leonardo robots after which have them play tennis or badminton.”