Tag Archives: encounter

#438006 Smellicopter Drone Uses Live Moth ...

Research into robotic sensing has, understandably I guess, been very human-centric. Most of us navigate and experience the world visually and in 3D, so robots tend to get covered with things like cameras and lidar. Touch is important to us, as is sound, so robots are getting pretty good with understanding tactile and auditory information, too. Smell, though? In most cases, smell doesn’t convey nearly as much information for us, so while it hasn’t exactly been ignored in robotics, it certainly isn’t the sensing modality of choice in most cases.

Part of the problem with smell sensing is that we just don’t have a good way of doing it, from a technical perspective. This has been a challenge for a long time, and it’s why we either bribe or trick animals like dogs, rats, vultures, and other animals to be our sensing systems for airborne chemicals. If only they’d do exactly what we wanted them to do all the time, this would be fine, but they don’t, so it’s not.

Until we get better at making chemical sensors, leveraging biology is the best we can do, and what would be ideal would be some sort of robot-animal hybrid cyborg thing. We’ve seen some attempts at remote controlled insects, but as it turns out, you can simplify things if you don’t use the entire insect, but instead just find a way to use its sensing system. Enter the Smellicopter.

There’s honestly not too much to say about the drone itself. It’s an open-source drone project called Crazyflie 2.0, with some additional off the shelf sensors for obstacle avoidance and stabilization. The interesting bits are a couple of passive fins that keep the drone pointed into the wind, and then the sensor, called an electroantennogram.

Image: UW

The drone’s sensor, called an electroantennogram, consists of a “single excised antenna” from a Manduca sexta hawkmoth and a custom signal processing circuit.

To make one of these sensors, you just, uh, “harvest” an antenna from a live hawkmoth. Obligingly, the moth antenna is hollow, meaning that you can stick electrodes up it. Whenever the olfactory neurons in the antenna (which is still technically alive even though it’s not attached to the moth anymore) encounter an odor that they’re looking for, they produce an electrical signal that the electrodes pick up. Plug the other ends of the electrodes into a voltage amplifier and filter, run it through an analog to digital converter, and you’ve got a chemical sensor that weighs just 1.5 gram and consumes only 2.7 mW of power. It’s significantly more sensitive than a conventional metal-oxide odor sensor, in a much smaller and more efficient form factor, making it ideal for drones.

To localize an odor, the Smellicopter uses a simple bioinspired approach called crosswind casting, which involves moving laterally left and right and then forward when an odor is detected. Here’s how it works:

The vehicle takes off to a height of 40 cm and then hovers for ten seconds to allow it time to orient upwind. The smellicopter starts casting left and right crosswind. When a volatile chemical is detected, the smellicopter will surge 25 cm upwind, and then resume casting. As long as the wind direction is fairly consistent, this strategy will bring the insect or robot increasingly closer to a singular source with each surge.

Since odors are airborne, they need a bit of a breeze to spread very far, and the Smellicopter won’t be able to detect them unless it’s downwind of the source. But, that’s just how odors work— even if you’re right next to the source, if the wind is blowing from you towards the source rather than the other way around, you might not catch a whiff of it.

Whenever the olfactory neurons in the antenna encounter an odor that they’re looking for, they produce an electrical signal that the electrodes pick up

There are a few other constraints to keep in mind with this sensor as well. First, rather than detecting something useful (like explosives), it’s going to detect the smells of pretty flowers, because moths like pretty flowers. Second, the antenna will literally go dead on you within a couple hours, since it only functions while its tissues are alive and metaphorically kicking. Interestingly, it may be possible to use CRISPR-based genetic modification to breed moths with antennae that do respond to useful smells, which would be a neat trick, and we asked the researchers—Melanie Anderson, a doctoral student of mechanical engineering at the University of Washington, in Seattle; Thomas Daniel, a UW professor of biology; and Sawyer Fuller, a UW assistant professor of mechanical engineering—about this, along with some other burning questions, via email.

IEEE Spectrum, asking the important questions first: So who came up with “Smellicopter”?

Melanie Anderson: Tom Daniel coined the term “Smellicopter”. Another runner up was “OdorRotor”!

In general, how much better are moths at odor localization than robots?

Melanie Anderson: Moths are excellent at odor detection and odor localization and need to be in order to find mates and food. Their antennae are much more sensitive and specialized than any portable man-made odor sensor. We can't ask the moths how exactly they search for odors so well, but being able to have the odor sensitivity of a moth on a flying platform is a big step in that direction.

Tom Daniel: Our best estimate is that they outperform robotic sensing by at least three orders of magnitude.

How does the localization behavior of the Smellicopter compare to that of a real moth?

Anderson: The cast-and-surge odor search strategy is a simplified version of what we believe the moth (and many other odor searching animals) are doing. It is a reactive strategy that relies on the knowledge that if you detect odor, you can assume that the source is somewhere up-wind of you. When you detect odor, you simply move upwind, and when you lose the odor signal you cast in a cross-wind direction until you regain the signal.

Can you elaborate on the potential for CRISPR to be able to engineer moths for the detection of specific chemicals?

Anderson: CRISPR is already currently being used to modify the odor detection pathways in moth species. It is one of our future efforts to specifically use this to make the antennae sensitive to other chemicals of interest, such as the chemical scent of explosives.

Sawyer Fuller: We think that one of the strengths of using a moth's antenna, in addition to its speed, is that it may provide a path to both high chemical specificity as well as high sensitivity. By expressing a preponderance of only one or a few chemosensors, we are anticipating that a moth antenna will give a strong response only to that chemical. There are several efforts underway in other research groups to make such specific, sensitive chemical detectors. Chemical sensing is an area where biology exceeds man-made systems in terms of efficiency, small size, and sensitivity. So that's why we think that the approach of trying to leverage biological machinery that already exists has some merit.

You mention that the antennae lifespan can be extended for a few days with ice- how feasible do you think this technology is outside of a research context?

Anderson: The antennae can be stored in tiny vials in a standard refrigerator or just with an ice pack to extend their life to about a week. Additionally, the process for attaching the antenna to the electrical circuit is a teachable skill. It is definitely feasible outside of a research context.

Considering the trajectory that sensor development is on, how long do you think that this biological sensor system will outperform conventional alternatives?

Anderson: It's hard to speak toward what will happen in the future, but currently, the moth antenna still stands out among any commercially-available portable sensors.

There have been some experiments with cybernetic insects; what are the advantages and disadvantages of your approach, as opposed to (say) putting some sort of tracking system on a live moth?

Daniel: I was part of a cyber insect team a number of years ago. The challenge of such research is that the animal has natural reactions to attempts to steer or control it.

Anderson: While moths are better at odor tracking than robots currently, the advantage of the drone platform is that we have control over it. We can tell it to constrain the search to a certain area, and return after it finishes searching.

What can you tell us about the health, happiness, and overall wellfare of the moths in your experiments?

Anderson: The moths are cold anesthetized before the antennae are removed. They are then frozen so that they can be used for teaching purposes or in other research efforts.

What are you working on next?

Daniel: The four big efforts are (1) CRISPR modification, (2) experiments aimed at improving the longevity of the antennal preparation, (3) improved measurements of antennal electrical responses to odors combined with machine learning to see if we can classify different odors, and (4) flight in outdoor environments.

Fuller: The moth's antenna sensor gives us a new ability to sense with a much shorter latency than was previously possible with similarly-sized sensors (e.g. semiconductor sensors). What exactly a robot agent should do to best take advantage of this is an open question. In particular, I think the speed may help it to zero in on plume sources in complex environments much more quickly. Think of places like indoor settings with flow down hallways that splits out at doorways, and in industrial settings festooned with pipes and equipment. We know that it is possible to search out and find odors in such scenarios, as anybody who has had to contend with an outbreak of fruit flies can attest. It is also known that these animals respond very quickly to sudden changes in odor that is present in such turbulent, patchy plumes. Since it is hard to reduce such plumes to a simple model, we think that machine learning may provide insights into how to best take advantage of the improved temporal plume information we now have available.

Tom Daniel also points out that the relative simplicity of this project (now that the UW researchers have it all figured out, that is) means that even high school students could potentially get involved in it, even if it’s on a ground robot rather than a drone. All the details are in the paper that was just published in Bioinspiration & Biomimetics. Continue reading

Posted in Human Robots

#437905 New Deep Learning Method Helps Robots ...

One of the biggest things standing in the way of the robot revolution is their inability to adapt. That may be about to change though, thanks to a new approach that blends pre-learned skills on the fly to tackle new challenges.

Put a robot in a tightly-controlled environment and it can quickly surpass human performance at complex tasks, from building cars to playing table tennis. But throw these machines a curve ball and they’re in trouble—just check out this compilation of some of the world’s most advanced robots coming unstuck in the face of notoriously challenging obstacles like sand, steps, and doorways.

The reason robots tend to be so fragile is that the algorithms that control them are often manually designed. If they encounter a situation the designer didn’t think of, which is almost inevitable in the chaotic real world, then they simply don’t have the tools to react.

Rapid advances in AI have provided a potential workaround by letting robots learn how to carry out tasks instead of relying on hand-coded instructions. A particularly promising approach is deep reinforcement learning, where the robot interacts with its environment through a process of trial-and-error and is rewarded for carrying out the correct actions. Over many repetitions it can use this feedback to learn how to accomplish the task at hand.

But the approach requires huge amounts of data to solve even simple tasks. And most of the things we would want a robot to do are actually comprised of many smaller tasks—for instance, delivering a parcel involves learning how to pick an object up, how to walk, how to navigate, and how to pass an object to someone else, among other things.

Training all these sub-tasks simultaneously is hugely complex and far beyond the capabilities of most current AI systems, so many experiments so far have focused on narrow skills. Some have tried to train AI on multiple skills separately and then use an overarching system to flip between these expert sub-systems, but these approaches still can’t adapt to completely new challenges.

Building off this research, though, scientists have now created a new AI system that can blend together expert sub-systems specialized for a specific task. In a paper in Science Robotics, they explain how this allows a four-legged robot to improvise new skills and adapt to unfamiliar challenges in real time.

The technique, dubbed multi-expert learning architecture (MELA), relies on a two-stage training approach. First the researchers used a computer simulation to train two neural networks to carry out two separate tasks: trotting and recovering from a fall.

They then used the models these two networks learned as seeds for eight other neural networks specialized for more specific motor skills, like rolling over or turning left or right. The eight “expert networks” were trained simultaneously along with a “gating network,” which learns how to combine these experts to solve challenges.

Because the gating network synthesizes the expert networks rather than switching them on sequentially, MELA is able to come up with blends of different experts that allow it to tackle problems none could solve alone.

The authors liken the approach to training people in how to play soccer. You start out by getting them to do drills on individual skills like dribbling, passing, or shooting. Once they’ve mastered those, they can then intelligently combine them to deal with more dynamic situations in a real game.

After training the algorithm in simulation, the researchers uploaded it to a four-legged robot and subjected it to a battery of tests, both indoors and outdoors. The robot was able to adapt quickly to tricky surfaces like gravel or pebbles, and could quickly recover from being repeatedly pushed over before continuing on its way.

There’s still some way to go before the approach could be adapted for real-world commercially useful robots. For a start, MELA currently isn’t able to integrate visual perception or a sense of touch; it simply relies on feedback from the robot’s joints to tell it what’s going on around it. The more tasks you ask the robot to master, the more complex and time-consuming the training will get.

Nonetheless, the new approach points towards a promising way to make multi-skilled robots become more than the sum of their parts. As much fun as it is, it seems like laughing at compilations of clumsy robots may soon be a thing of the past.

Image Credit: Yang et al., Science Robotics Continue reading

Posted in Human Robots

#437845 Video Friday: Harmonic Bionics ...

Video Friday is your weekly selection of awesome robotics videos, collected by your Automaton bloggers. We’ll also be posting a weekly calendar of upcoming robotics events for the next few months; here's what we have so far (send us your events!):

ICRA 2020 – May 31-August 31, 2020 – [Virtual Conference]
RSS 2020 – July 12-16, 2020 – [Virtual Conference]
CLAWAR 2020 – August 24-26, 2020 – [Virtual Conference]
ICUAS 2020 – September 1-4, 2020 – Athens, Greece
ICRES 2020 – September 28-29, 2020 – Taipei, Taiwan
ICSR 2020 – November 14-16, 2020 – Golden, Colorado
Let us know if you have suggestions for next week, and enjoy today's videos.

Designed to protect employees and passengers from both harmful pathogens and cleaning agents, Breezy One can quickly, safely and effectively decontaminate spaces over 100,000 square feet in 1.5 hours with a patented, environmentally safe disinfectant. Breezy One was co-developed with the City of Albuquerque’s Aviation Department, where it autonomously sanitizes the Sunport’s facilities every night in the ongoing fight against COVID-19.

[ Fetch Robotics ]

Harmonic Bionics is redefining upper extremity neurorehabilitation with intelligent robotic technology designed to maximize patient recovery. Harmony SHR, our flagship product, works with a patient’s scapulohumeral rhythm (SHR) to enable natural, comprehensive therapy for both arms. When combined with Harmony’s Weight Support mode, this unique shoulder design may allow for earlier initiation of post-stroke therapy as Harmony can support a partial dislocation or subluxation of the shoulder prior to initiating traditional therapy exercises.

Harmony's Preprogrammed Exercises promotes functional treatment through patient-specific movements that can enable an increased number of repetitions per session without placing a larger physical burden on therapists or their resources. As the only rehabilitation exoskeleton with Bilateral Sync Therapy (BST), Harmony enables intent-based therapy by registering healthy arm movements and synchronizing that motion onto the stroke-affected side to help reestablish neural pathways.

[ Harmonic Bionics ]

Thanks Mok!

Some impressive work here from IHMC and IIT getting Atlas to take steps upward in a way that’s much more human-like than robot-like, which ends up reducing maximum torque requirements by 20 percent.

[ Paper ]

GITAI’s G1 is the space dedicated general-purpose robot. G1 robot will enable automation of various tasks internally & externally on space stations and for lunar base development.

[ GITAI ]

Malloy Aeronautics, which now makes drones rather than hoverbikes, has been working with the Royal Navy in New Zealand to figure out how to get cargo drones to land on ships.

The challenge was to test autonomous landing of heavy lift UAVs on a moving ship, however, due to the Covid19 lockdown no ship trails were possible. The moving deck was simulated by driving a vehicle and trailer across an airfield while carrying out multiple landing and take-offs. The autonomous system partner was Planck Aerosystems and autolanding was triggered by a camera on the UAV reading a QR code on the trailer.

[ Malloy Aeronautics ]

Thanks Paul!

Tertill looks to be relentlessly effective.

[ Franklin Robotics ]

A Swedish company, TikiSafety has experienced a record amount of orders for their protective masks. At ABB, we are grateful for the opportunity to help Tiki Safety to speed up their manufacturing process from 6 minutes to 40 seconds.

[ Tiki Safety ]

The Korea Atomic Energy Research Institute is not messing around with ARMstrong, their robot for nuclear and radiation emergency response.

[ KAERI ]

OMOY is a robot that communicates with its users via internal weight shifting.

[ Paper ]

Now this, this is some weird stuff.

[ Segway ]

CaTARo is a Care Training Assistant Robot from the AIS Lab at Ritsumeikan University.

[ AIS Lab ]

Originally launched in 2015 to assist workers in lightweight assembly tasks, ABB’s collaborative YuMi robot has gone on to blaze a trail in a raft of diverse applications and industries, opening new opportunities and helping to fire people’s imaginations about what can be achieved with robotic automation.

[ ABB ]

This music video features COMAN+, from the Humanoids and Human Centered Mechatronics Lab at IIT, doing what you’d call dance moves if you dance like I do.

[ Alex Braga ] via [ IIT ]

The NVIDIA Isaac Software Development Kit (SDK) enables accelerated AI robot development workflows. Stacked with new tools and application support, Isaac SDK 2020.1 is an end-to-end solution supporting each step of robot fleet deployment, from design collaboration and training to the ongoing maintenance of AI applications.

[ NVIDIA ]

Robot Spy Komodo Dragon and Spy Pig film “a tender moment” between Komodo dragons but will they both survive the encounter?

[ BBC ] via [ Laughing Squid ]

This is part one of a mostly excellent five-part documentary about ROS produced by Red Hat. I say mostly only because they put ME in it for some reason, but fortunately, they talked with many of the core team that developed ROS back at Willow Garage back in the day, and it’s definitely worth watching.

[ Red Hat Open Source Stories ]

It’s been a while, but here’s an update on SRI’s Abacus Drive, from Alexander Kernbaum.

[ SRI ]

This Robots For Infectious Diseases interview features IEEE Fellow Antonio Bicchi, professor of robotics at the University of Pisa, talking about how Italy has been using technology to help manage COVID-19.

[ R4ID ]

Two more interviews this week of celebrity roboticists from MassRobotics: Helen Greiner and Marc Raibert. I’d introduce them, but you know who they are already!

[ MassRobotics ] Continue reading

Posted in Human Robots

#437630 How Toyota Research Envisions the Future ...

Yesterday, the Toyota Research Institute (TRI) showed off some of the projects that it’s been working on recently, including a ceiling-mounted robot that could one day help us with household chores. That system is just one example of how TRI envisions the future of robotics and artificial intelligence. As TRI CEO Gill Pratt told us, the company is focusing on robotics and AI technology for “amplifying, rather than replacing, human beings.” In other words, Toyota wants to develop robots not for convenience or to do our jobs for us, but rather to allow people to continue to live and work independently even as we age.

To better understand Toyota’s vision of robotics 15 to 20 years from now, it’s worth watching the 20-minute video below, which depicts various scenarios “where the application of robotic capabilities is enabling members of an aging society to live full and independent lives in spite of the challenges that getting older brings.” It’s a long video, but it helps explains TRI’s perspective on how robots will collaborate with humans in our daily lives over the next couple of decades.

Those are some interesting conceptual telepresence-controlled bipeds they’ve got running around in that video, right?

For more details, we sent TRI some questions on how it plans to go from concepts like the ones shown in the video to real products that can be deployed in human environments. Below are answers from TRI CEO Gill Pratt, who is also chief scientist for Toyota Motor Corp.; Steffi Paepcke, senior UX designer at TRI; and Max Bajracharya, VP of robotics at TRI.

IEEE Spectrum: TRI seems to have a more explicit focus on eventual commercialization than most of the robotics research that we cover. At what point TRI starts to think about things like reliability and cost?

Photo: TRI

Toyota is exploring robots capable of manipulating dishes in a sink and a dishwasher, performing experiments and simulations to make sure that the robots can handle a wide range of conditions.

Gill Pratt: It’s a really interesting question, because the normal way to think about this would be to say, well, both reliability and cost are product development tasks. But actually, we need to think about it at the earliest possible stage with research as well. The hardware that we use in the laboratory for doing experiments, we don’t worry about cost there, or not nearly as much as you’d worry about for a product. However, in terms of what research we do, we very much have to think about, is it possible (if the research is successful) for it to end up in a product that has a reasonable cost. Because if a customer can’t afford what we come up with, maybe it has some academic value but it’s not actually going to make a difference in their quality of life in the real world. So we think about cost very much from the beginning.

The same is true with reliability. Right now, we’re working very hard to make our control techniques robust to wide variations in the environment. For instance, in work that Russ Tedrake is doing with manipulating dishes in a sink and a dishwasher, both in physical testing and in simulation, we’re doing thousands and now millions of different experiments to make sure that we can handle the edge cases and it works over a very wide range of conditions.

A tremendous amount of work that we do is trying to bring robotics out of the age of doing demonstrations. There’s been a history of robotics where for some time, things have not been reliable, so we’d catch the robot succeeding just once and then show that video to the world, and people would get the mis-impression that it worked all of the time. Some researchers have been very good about showing the blooper reel too, to show that some of the time, robots don’t work.

“A tremendous amount of work that we do is trying to bring robotics out of the age of doing demonstrations. There’s been a history of robotics where for some time, things have not been reliable, so we’d catch the robot succeeding just once and then show that video to the world, and people would get the mis-impression that it worked all of the time.”
—Gill Pratt, TRI

In the spirit of sharing things that didn’t work, can you tell us a bit about some of the robots that TRI has had under development that didn’t make it into the demo yesterday because they were abandoned along the way?

Steffi Paepcke: We’re really looking at how we can connect people; it can be hard to stay in touch and see our loved ones as much as we would like to. There have been a few prototypes that we’ve worked on that had to be put on the shelf, at least for the time being. We were exploring how to use light so that people could be ambiently aware of one another across distances. I was very excited about that—the internal name was “glowing orb.” For a variety of reasons, it didn’t work out, but it was really fascinating to investigate different modalities for keeping in touch.

Another prototype we worked on—we found through our research that grocery shopping is obviously an important part of life, and for a lot of older adults, it’s not necessarily the right answer to always have groceries delivered. Getting up and getting out of the house keeps you physically active, and a lot of people prefer to continue doing it themselves. But it can be challenging, especially if you’re purchasing heavy items that you need to transport. We had a prototype that assisted with grocery shopping, but when we pivoted our focus to Japan, we found that the inside of a Japanese home really needs to stay inside, and the outside needs to stay outside, so a robot that traverses both domains is probably not the right fit for a Japanese audience, and those were some really valuable lessons for us.

Photo: TRI

Toyota recently demonstrated a gantry robot that would hang from the ceiling to perform tasks like wiping surfaces and clearing clutter.

I love that TRI is exploring things like the gantry robot both in terms of near-term research and as part of its long-term vision, but is a robot like this actually worth pursuing? Or more generally, what’s the right way to compromise between making an environment robot friendly, and asking humans to make changes to their homes?

Max Bajracharya: We think a lot about the problems that we’re trying to address in a holistic way. We don’t want to just give people a robot, and assume that they’re not going to change anything about their lifestyle. We have a lot of evidence from people who use automated vacuum cleaners that people will adapt to the tools you give them, and they’ll change their lifestyle. So we want to think about what is that trade between changing the environment, and giving people robotic assistance and tools.

We certainly think that there are ways to make the gantry system plausible. The one you saw today is obviously a prototype and does require significant infrastructure. If we’re going to retrofit a home, that isn’t going to be the way to do it. But we still feel like we’re very much in the prototype phase, where we’re trying to understand whether this is worth it to be able to bypass navigation challenges, and coming up with the pros and cons of the gantry system. We’re evaluating whether we think this is the right approach to solving the problem.

To what extent do you think humans should be either directly or indirectly in the loop with home and service robots?

Bajracharya: Our goal is to amplify people, so achieving this is going to require robots to be in a loop with people in some form. One thing we have learned is that using people in a slow loop with robots, such as teaching them or helping them when they make mistakes, gives a robot an important advantage over one that has to do everything perfectly 100 percent of the time. In unstructured human environments, robots are going to encounter corner cases, and are going to need to learn to adapt. People will likely play an important role in helping the robots learn. Continue reading

Posted in Human Robots

#437624 AI-Powered Drone Learns Extreme ...

Quadrotors are among the most agile and dynamic machines ever created. In the hands of a skilled human pilot, they can do some astonishing series of maneuvers. And while autonomous flying robots have been getting better at flying dynamically in real-world environments, they still haven’t demonstrated the same level of agility of manually piloted ones.

Now researchers from the Robotics and Perception Group at the University of Zurich and ETH Zurich, in collaboration with Intel, have developed a neural network training method that “enables an autonomous quadrotor to fly extreme acrobatic maneuvers with only onboard sensing and computation.” Extreme.

There are two notable things here: First, the quadrotor can do these extreme acrobatics outdoors without any kind of external camera or motion-tracking system to help it out (all sensing and computing is onboard). Second, all of the AI training is done in simulation, without the need for an additional simulation-to-real-world (what researchers call “sim-to-real”) transfer step. Usually, a sim-to-real transfer step means putting your quadrotor into one of those aforementioned external tracking systems, so that it doesn’t completely bork itself while trying to reconcile the differences between the simulated world and the real world, where, as the researchers wrote in a paper describing their system, “even tiny mistakes can result in catastrophic outcomes.”

To enable “zero-shot” sim-to-real transfer, the neural net training in simulation uses an expert controller that knows exactly what’s going on to teach a “student controller” that has much less perfect knowledge. That is, the simulated sensory input that the student ends up using as it learns to follow the expert has been abstracted to present the kind of imperfect, imprecise data it’s going to encounter in the real world. This can involve things like abstracting away the image part of the simulation until you’d have no way of telling the difference between abstracted simulation and abstracted reality, which is what allows the system to make that sim-to-real leap.

The simulation environment that the researchers used was Gazebo, slightly modified to better simulate quadrotor physics. Meanwhile, over in reality, a custom 1.5-kilogram quadrotor with a 4:1 thrust to weight ratio performed the physical experiments, using only a Nvidia Jetson TX2 computing board and an Intel RealSense T265, a dual fisheye camera module optimized for V-SLAM. To challenge the learning system, it was trained to perform three acrobatic maneuvers plus a combo of all of them:

Image: University of Zurich/ETH Zurich/Intel

Reference trajectories for acrobatic maneuvers. Top row, from left: Power Loop, Barrel Roll, and Matty Flip. Bottom row: Combo.

All of these maneuvers require high accelerations of up to 3 g’s and careful control, and the Matty Flip is particularly challenging, at least for humans, because the whole thing is done while the drone is flying backwards. Still, after just a few hours of training in simulation, the drone was totally real-world competent at these tricks, and could even extrapolate a little bit to perform maneuvers that it was not explicitly trained on, like doing multiple loops in a row. Where humans still have the advantage over drones is (as you might expect since we’re talking about robots) is quickly reacting to novel or unexpected situations. And when you’re doing this sort of thing outdoors, novel and unexpected situations are everywhere, from a gust of wind to a jealous bird.

For more details, we spoke with Antonio Loquercio from the University of Zurich’s Robotics and Perception Group.

IEEE Spectrum: Can you explain how the abstraction layer interfaces with the simulated sensors to enable effective sim-to-real transfer?

Antonio Loquercio: The abstraction layer applies a specific function to the raw sensor information. Exactly the same function is applied to the real and simulated sensors. The result of the function, which is “abstracted sensor measurements,” makes simulated and real observation of the same scene similar. For example, suppose we have a sequence of simulated and real images. We can very easily tell apart the real from the simulated ones given the difference in rendering. But if we apply the abstraction function of “feature tracks,” which are point correspondences in time, it becomes very difficult to tell which are the simulated and real feature tracks, since point correspondences are independent of the rendering. This applies for humans as well as for neural networks: Training policies on raw images gives low sim-to-real transfer (since images are too different between domains), while training on the abstracted images has high transfer abilities.

How useful is visual input from a camera like the Intel RealSense T265 for state estimation during such aggressive maneuvers? Would using an event camera substantially improve state estimation?

Our end-to-end controller does not require a state estimation module. It shares however some components with traditional state estimation pipelines, specifically the feature extractor and the inertial measurement unit (IMU) pre-processing and integration function. The input of the neural networks are feature tracks and integrated IMU measurements. When looking at images with low features (for example when the camera points to the sky), the neural net will mainly rely on IMU. When more features are available, the network uses to correct the accumulated drift from IMU. Overall, we noticed that for very short maneuvers IMU measurements were sufficient for the task. However, for longer ones, visual information was necessary to successfully address the IMU drift and complete the maneuver. Indeed, visual information reduces the odds of a crash by up to 30 percent in the longest maneuvers. We definitely think that event camera can improve even more the current approach since they could provide valuable visual information during high speed.

“The Matty Flip is probably one of the maneuvers that our approach can do very well … It is super challenging for humans, since they don’t see where they’re going and have problems in estimating their speed. For our approach the maneuver is no problem at all, since we can estimate forward velocities as well as backward velocities.”
—Antonio Loquercio, University of Zurich

You describe being able to train on “maneuvers that stretch the abilities of even expert human pilots.” What are some examples of acrobatics that your drones might be able to do that most human pilots would not be capable of?

The Matty Flip is probably one of the maneuvers that our approach can do very well, but human pilots find very challenging. It basically entails doing a high speed power loop by always looking backward. It is super challenging for humans, since they don’t see where they’re going and have problems in estimating their speed. For our approach the maneuver is no problem at all, since we can estimate forward velocities as well as backward velocities.

What are the limits to the performance of this system?

At the moment the main limitation is the maneuver duration. We never trained a controller that could perform maneuvers longer than 20 seconds. In the future, we plan to address this limitation and train general controllers which can fly in that agile way for significantly longer with relatively small drift. In this way, we could start being competitive against human pilots in drone racing competitions.

Can you talk about how the techniques developed here could be applied beyond drone acrobatics?

The current approach allows us to do acrobatics and agile flight in free space. We are now working to perform agile flight in cluttered environments, which requires a higher degree of understanding of the surrounding with respect to this project. Drone acrobatics is of course only an example application. We selected it because it makes a stress test of the controller performance. However, several other applications which require fast and agile flight can benefit from our approach. Examples are delivery (we want our Amazon packets always faster, don’t we?), search and rescue, or inspection. Going faster allows us to cover more space in less time, saving battery costs. Indeed, agile flight has very similar battery consumption of slow hovering for an autonomous drone.

“Deep Drone Acrobatics,” by Elia Kaufmann, Antonio Loquercio, René Ranftl, Matthias Müller, Vladlen Koltun, and Davide Scaramuzza from the Robotics and Perception Group at the University of Zurich and ETH Zurich, and Intel’s Intelligent Systems Lab, was presented at RSS 2020. Continue reading

Posted in Human Robots