Morgan Pope is a Ph.D student investigating robots that live at the boundary of airborne and surface locomotion at Stanford’s Biomimetics and Dextrous Manipulation Lab. He’s the lead author on a paper about SCAMP that is in review for IEEE Transactions on Robotics, and enjoys reading, Star Wars, and trying to keep up with his three small children.
What goes up must come down—unless it can perch on something first. Quadrotors have limited endurance because of restrictions on battery capacity and the physics of small-scale flight, but perching can allow them to operate for hours or even days, gathering data or performing communication tasks while stationary. Perching can be tricky, because the odds of your drone landing in just the right place are low. Adding the ability to climb allows your drone to reposition itself more accurately, with the added bonus that it works if it’s too windy for flight.
At Stanford’s Biomimetics and Dexterous Manipulation Lab, we saw a chance to combine our experience in perching and climbing with a new robot capable of multi-modal operation in unstructured outdoor environments. The result was the Stanford Climbing and Aerial Maneuvering Platform, a collection of words that gives us an excuse to call our robot SCAMP.
SCAMP is the first robot to combine flying, perching with passive attachment technology, and climbing. It can also recover from climbing failures, as well as take off when it’s ready to fly again. It does all of this outdoors, using only onboard sensing and computation, leveraging lessons from all our previous climbing robots, our recent work in perching, and mother nature.
Over the last decade, we have developed a series of climbing robots that use directional adhesives to create smooth, reliable climbing gaits. Microspines (hardened steel barbs mounted on compliant suspensions) are our original directional adhesive, and they enabled Spinybot and RiSE to climb rough surfaces. We then developed gecko adhesive, which enabled directional adhesion on glass and led to the creation of Stickybot I, II, and III. As we refined our understanding of directional adhesion and climbing, we were able to miniaturize Stickybot into a 9-gram robot capable of pulling 100x its body weight up a wall. This was made possible by using a compact, powerful servo and alternating loads between two feet which move on simple, one-dimensional trajectories.
We took these lessons and re-applied them in the context of microspines to create SCAMP’s climbing mechanism. We used the same compact, powerful servo, and the same strategy of transferring loads between two feet, but since we were looking for maneuverability instead of carrying capacity, we designed it to have a longer stroke length (9 cm/step instead of 1.2 cm/step). We also learned that it helped to add motion towards and away from the wall because concrete and stucco are not as flat and predictable as a glass window. The end result is a climbing mechanism that uses one high torque-density servo to drive long steps up the wall, and one even smaller servo to actuate motion towards and away from the wall. These two servos, combined with the carbon fiber frame and spiny feet, weigh only 11 grams. In effect, we’ve taken our 9-gram micro glass climber, modified it for speed instead of load capability, given it an extra servo to handle two-dimensional surface profiles, outfitted it with microspines, and strapped it to a tiny quadrotor.
The leg design of SCAMP is reminiscent of many climbing insects, from daddy longlegs to the praying mantis, and that’s no accident. Animals want long, efficient steps, but are limited by the weight of their limbs. As we descend into the realm of insects, allometric scaling laws mean long, thin, almost weightless legs become the preferred solution. SCAMP isn’t quite insect-sized, but the robot is small enough that modern engineered materials like carbon fiber and Spectra let us create legs that are as long and weight-efficient as a those of a climbing insect.
Flying and Perching
A good climbing mechanism is not the optimal perching mechanism, and that means we had to reconsider our approach to attaching to the wall. Previously, we’ve perched outdoors using a fixed-wing robot, and indoors using quadrotors and a motion-capture system. In both cases, the vehicles lost control authority as the maneuver neared completion, meaning that we had to engineer a suspension to absorb the impact and, in the quadrotor case, an opposed-grip attachment system that could resist loads in any direction. In the natural world, animals use aerodynamic forces throughout the perching process, and that’s the approach we took for SCAMP.
We put SCAMP’s climbing mechanism on top of the quadrotor, so that the rotors could actively press the robot into the wall. When we combined that with a long tail acting as a pivot point, we had a system that was able to reliably push itself onto the wall using aerodynamic forces.
To perch, we fly tail-first towards a wall until we detect an impact using our onboard accelerometers. Then we turn the rotor thrust to maximum. The geometry of the problem pretty much guarantees that SCAMP will end up with its climbing gear pressed against the target surface. The rotors keep SCAMP’s feet in contact until the vibrations of impact have a chance to subside. Then, as microspines engage, we can turn the rotors off and begin climbing.
Synergies of Multi-Modal Operation
Having to retool our perching strategy is one example of how putting two very different modes of locomotion together requires a certain amount of compromise. However, we found that these compromises were offset by some interesting new synergies between our flying and climbing modes.
While SCAMP’s rotors initially seemed like dead weight for the climbing gear to pull up the wall, we soon learned that they could make a big difference in our locomotion along a surface. When SCAMP misses its grip and starts to fall, it notices the sudden vertical acceleration and turns its rotors on briefly. This action pushes it back onto the wall so it can resume climbing, and we can save ourselves a trip to the robot infirmary. Nature figured this out first, of course—even flightless animals can use aerodynamic forces to recover from a climbing failure, like the rainforest ants studied by Yanoviak, Dudley, and Kaspari, which use their legs and torsos to steer themselves back to the nearest trunk after a fall.
Rotors can also be used to help SCAMP’s microspines engage with the wall. For SCAMP, one of the biggest challenges of climbing a vertical surface is that the center of mass of the robot is inevitably cantilevered away from the wall, which means that the feet need to generate some adhesive force to keep the robot from pitching backward. However, if we turn SCAMP’s rotors on just a bit, we can compensate for this pitch-back moment aerodynamically, which makes the job of the feet significantly easier. Here again nature beat us to the punch, most dramatically in the case of the Chukar partridge, which uses its wings to give itself the ability to run up vertical posts that it would never be able to scale otherwise. Natural analogs like these are easy to find for SCAMP: animals are inherently multi-modal, and a huge variety of animals, from flying squirrels to woodpeckers, alternate between flying, perching, and climbing to optimize their arboreal locomotion strategy.
The Power of Flying and Climbing
Extended mission life is the main motivation for perching, so we hooked SCAMP up to a current meter to figure out exactly how much longer it could operate thanks to its ability to hang passively on a wall. SCAMP’s microprocessor is a little power-hungry, drawing more than a 100 milliamps just passing the time (a more efficient model can easily operate on two orders of magnitude less power, but the quadrotor electronics we used weren’t optimized for power draw because it makes little difference compared to the multiple amps required for flying). Even so, dedicated perching can easily extend mission life from 3 minutes to 2 hours.
Flying is currently a more energy efficient method of travel for SCAMP than climbing. That’s more of a function of some stickiness in our servo gear train and our power-hungry microprocessor than any fundamental law. In fact, one would expect climbing to be more efficient in general, because you don’t have to do the extra work of accelerating air molecules as you do in flight. Whatever the final efficiency turns out to be, perching and climbing is much more precise and reliable than perching alone, especially if you want your robot to be somewhere exact—for the right vantage point or a better signal, for instance. It also means that repositioning yourself doesn’t require the more hazardous maneuvers of taking off and re-perching. A good analogy would be between flying and driving: a modern wide-body passenger jet gets an efficiency of roughly 100 miles per gallon per passenger, but you’ll run into problems if you try and use it to get from your house to the grocery store. The gains in efficiency are offset by the loss in precision and by the costs of takeoff and landing, just as in the case of a perching robot.
Future Work: Adaptive Gait and the SCAMP Family
Looking forward, there are many things to do with SCAMP that we haven’t had time to implement yet. For example, we can tune SCAMP’s gait using the two servos, to compensate for surface roughness and slipperiness. For slippery surfaces , a shorter stroke length can be more reliable because there’s less of a change in pitch-back moment over the course of a stroke. For surfaces that are relatively smooth and have mainly negative “pits” for footholds, keeping the feet on a very flat trajectory maximizes our chance of finding a good grip, especially if the footholds are few and far between. Conversely, for rougher surfaces with lots of bumps, it’s better to limit the time when both feet are moving near the wall. In future work, we’d like to implement adaptive gait control for SCAMP—algorithms that allow the robot to react to failure by dynamically changing its climbing strategy. This is a place where machine learning might be a powerful real-time tool, and the right algorithm might even generate effective gaits that we hadn’t thought of before.
We also see SCAMP as the starting place for an entire family of perching and climbing robots of varying scales and attachment strategies. The lessons we learned from SCAMP should allow us to tackle new surfaces, new environments, and different quadrotor platforms with new sensing and communication abilities. In the meantime, SCAMP will continue doing its best to defy gravity and go up without coming down.
[ SCAMP ]