Diligent Robotics, founded by Andrea Thomaz and Vivian Chu, develops socially intelligent automation solutions for hospitals. Moxi, their flagship robot, delivers items like medications and wound dressings between departments to save the clinical staff’s time. Diligent has just closed their Series B funding round with $30 million. 

We sat down with Dr. Thomaz to talk about Moxi, how to manage people’s expectations about robots, and advice for young people and women in robotics. This interview has been lightly edited for clarity. 

Andrea Thomaz with Diligent’s flagship robot Moxi.

What kinds of problems are you trying to solve with Moxi?

We are building Moxi to help hospitals with the massive workforce shortage that they’re seeing now more than ever. We actually started the company with the same intention several years ago before there was a worldwide pandemic, and it really has just gotten to be an even bigger problem for hospitals. I feel really strongly that robots have a place to play in teamwork environments, and hospitals are a great example of that. There’s no one person’s job in a hospital that you would actually want to give over to automation or robots, but there are tiny little bits of a lot of people’s jobs that are absolutely able to be automated and that we can give over to delivery robots in particular like Moxi. [The main problem we’re trying to solve with Moxi is] point to point delivery, where we’re fetching and gathering things and taking them from one area of the hospital to another. 

Hospitals have a lot of stuff that’s moving around every day. Every person in the hospital is going to have certain medications that need to be delivered to them, certain lab samples that need to be taken and delivered to the central lab, certain supplies that need to come up to them, food and nutrition every day. You have a lot of stuff that’s coming and going between patient units and all these different support departments. 

Every one of these support departments has a process in place for getting the stuff moved around, but no matter what, there’s stuff that happens every single day that requires ad-hoc [deliveries] to happen between all of these departments and different nursing units. So sometimes that’s going to be a nurse that just needs to get something for their patient and they want that to happen as soon as possible. They’re trying to discharge their patient, they need a particular wound dressing kit, they’re going to run down and get it because they want to help their patient get out. Or if there’s something that needs to be hand carried because the regular rounding of medications has already happened, a lot of times you’ll have a pharmacy technician stop what they’re doing and go and run some infusion meds for a cancer patient, for example. It sort of falls between these departments. There’s different people that would be involved but a lot of times it does fall on the nursing units themselves. A nurse explained to us one time that nurses are the last line of defense in patient care.

Moxi performing a delivery for a clinical staff member.

What is changing with this most recent round of funding?

Over the last 6-12 months, the demand has really skyrocketed such that we’re barely keeping up with the demand for people wanting to implement robots in their hospitals. That’s the reason why we’re raising this round of funding, expanding the team, and expanding our ability to capitalize on that demand. A couple of years ago, if we were working with a hospital it was because they had some special funds set aside for innovation or they had a CTO or a CIO that had a background in robotics, but it certainly wasn’t the first thing that every hospital CIO was thinking about. Now that has completely changed. We’re getting cold outreach on our website from CIOs of hospitals saying “I need to develop a robotic strategy for our hospital and I want to learn about your solution.” Through the pandemic, I think everyone has seen that the workforce shortage in hospitals is only getting worse in the near term. Everybody wants to plan for the future and do everything they can to take small tasks off of the plates of their clinical teams. It’s been really exciting to be part of that market change and see that shift to where everybody is really really open to automation. Before we had to say “No no no, this is not the future, I promise it’s not scifi, I promise these really work.” Now [the climate has] really shifted to people understanding “This is actually something that can impact my teams.”

[Two of our investors are hospitals, and] that’s been one of our most exciting parts of this round. It’s always great to have a successful funding round, but to have strategic partners like Cedars-Sinai and Shannon Healthcare coming in and saying “Yeah, we actually want to build this alongside you” — it’s pretty exciting to have customers like that. 

What kinds of technical problems did you run into when you were either building Moxi or deploying it in a hospital environment? How did you solve those problems? 

One that was almost surprising in how often it came up, and really impacted our ability [to run Moxi in the hospital environment] because we have a software-based robotic solution that is connecting at a regular basis to cloud services, [was that] we had no clue how terrible hospital WiFi was going to be. We actually spent quite a while building in backup systems to be able to use WiFi, backup to LTE if we have to, but be smart about that so we’re not spending a whole bunch of money on LTE data. That was a problem that seemed very specific to hospitals in particular.

Another one was security and compliance. We just didn’t know what some of the different requirements were for hospitals until we actually got into the environments and started interacting with customers and understanding what they wanted to use Moxi for. When we were first doing research trials in 2018 or 2019, we had a version of the robot that was a little bit different than the one we have today. It had lots of open containers so you could just put whatever you wanted to on the robot and send it over to another location. We quickly learned that that limited what the robot was allowed to carry, because so much of what [the customers] wanted was to understand who pulled something out of the robot. So now we have an RF badge reader on the robot that is connected to locking storage containers that are only going to open if you’re the kind of person that is allowed to open the robot. That was an interesting technical challenge that we didn’t know about until after we got out there. 

Moxi’s locking storage containers.

How did you work with nurses and the other healthcare professionals you were working with to figure out what would be the most helpful robot for them? 

My background, and my co-founder Vivian Chu’s background, is in human-robot interaction so we knew that we didn’t know enough about nursing or the hospital environment. We spent the first 9 months of the company in 2018 building out our research prototype. It looked a lot like what Moxi looks like today. Under the hood it was completely different than what we have today in terms of the reliability and robustness of the hardware and software, but it was enough to get that platform out and have it deployed with nursing units. We embedded ourselves with four different nursing units across Texas over a year-long period. We would spend about 6-8 weeks with a nursing department, and we were just there — engineers, product people, and everybody in the company was cycling in and out a week or two at a time. 

We would ask those nurses: “What would you actually want a robot like this to do?” Part of this that was really important was they didn’t have good ideas about what they would want the robot to do until they saw the robot. It was a very participatory design, where they had to see and get a sense for the robot before they would have good ideas of what they would want the robot to do. Then we would take those ideas [to the company] and come back and say “Yes we can do that,” or “No we can’t do that.” We came out of that whole process with a really great idea. We like to say that’s where we found our product market fit — that’s where we really understood that what was going to be most valuable for the robot to do was connecting the nursing units to these other departments. We can help a nurse with supply management and getting things from place to place within their department, or we can help them with things that are coming from really far away. [The second one] was actually impacting their time way way more.

Because the capabilities of robotic systems are usually misinterpreted, it can be really hard to manage the relationship with stakeholders and customers and set appropriate expectations. How did you manage that relationship?

We do a lot of demonstrations, but still with almost every single implementation you get questions about some robot in Hollywood, [and you have to say] “No, that’s the movies” and explain exactly what Moxi does. 

From a design perspective, we also limit the English phrases that come out of Moxi’s mouth just because we don’t want to communicate a really high level of intelligence. There are lots of canned phrases and interactions on the iPad instead of via voice, and a lot of times the robot will just make meeps and beeps and flash lights and things like that. 

Before starting the company, I had a lab, and one of the big research topics that we had for a number of years was embodied dialogue — how robots could have a real conversation with people. I had a very good appreciation for how hard that problem is, and also for just how much people want it. People come up to a robot, and they want it to be able to talk to them. How you can [set expectations] with the design and behavior of the robot has been a focus of mine since before we started the company. We purposefully don’t make the robot look very human-like because we don’t want there to be android human-level expectations, but [the robot does have a face and eyes so it can] communicate “I’m looking at that thing” and “I’m about to manipulate that thing,” which we think is important. It’s really about striking that balance. 

What would you say is one lesson that you’ve learned from your work at Diligent so far and how are you looking to apply this lesson moving forward?

The difference between research and practice. On the one hand, the motivation and reason for starting a company is that you want to see the kinds of things that you’ve done in the research lab really make it out into the world and start to impact real people and their work. That’s been one of the most fascinating, impactful, and inspiring things about starting Diligent: Being able to go and see nurses when Moxi is doing work for them. They are so thankful! If you just hang back and watch Moxi come and do a delivery, almost always people are super excited to see the robot. They get their delivery and they’re like, “Oh, thank you Moxi!” That feels like we’re really making a difference in a way that you just don’t get with just research contributions that don’t make it all the way out into the world. 

That being said though, there is a long tail of things that you have to solve from an engineering perspective beyond [developing a feature]. My VP of engineering Starr Corbin has this great way of putting it: The research team will get a certain thing on the product to be feature complete, where we’ve demonstrated that this feature works and it’s a good solution, but then there’s this whole phase that has to happen after that to get the feature to be production ready. I would say my biggest lesson is probably everything that it takes, and the entire team of people it takes, to get something from being feature complete to production ready. I have a deep appreciation for that. How fast we can move things out into the world is really dictated by some of that. 

Andrea Thomaz (left) and Vivian Chu with Moxi.

What advice would you give young women in robotics? 

If I put my professor hat on, I always had advice that I liked to give women in robotics, in academia, and just kind of pursuing things in general. Imposter syndrome is real, and everybody feels it. All you can do to combat it is not underestimate yourself. Speak up and know that you deserve a seat at the table. It’s all about hard work, but also making sure that your voice is heard. Some of the mentorship that I gave to a lot of my women grad students when I was a professor was around speaking engagements, speaking styles, and communication. It can be really uncomfortable when you’re the only anything in the room to stand up and feel like you deserve to be the one speaking, and so the more that you practice doing that, the more comfortable it can feel, the more confident you’ll feel in yourself and your voice. I think finding that confident voice is a really important skill that you have to develop early on in your career. 

What’s one piece of advice you’ve received that you always turn to when things are tough? 

There are two mentors that I’ve had who are women in AI and robotics. [In my] first year as a faculty member [the first mentor] came and gave a research seminar talk. I for some reason got to take her out to lunch by myself, so we had this amazing one-on-one. We talked a little bit about her talk, probably half of the lunch we talked about technical things, and then she just kind of turned the conversation [around] and said “Andrea, don’t forget to have a family.” Like, don’t forget to focus on that part of your life — it’s the most important thing. She got on a soapbox and said “You have to have a work life balance it’s so important. Don’t forget to focus on building a family for yourself, whatever that looks like.” That really stuck with me, especially as [when you’re] early in your career you’re worried about nothing but success. It was really powerful to have somebody strong and influential like that telling you “No, no, this is important and you need to focus on this.” 

The other person that’s always been an inspiration and mentor for me that I’ll highlight [was the professor teaching a class I TA’d for at MIT]. I had found a bug in one of her homework problems, and she was like, “Oh, fascinating.” She was so excited that I had found a question that she didn’t know the answer to. She [just said], “Oh my gosh I don’t know, let’s go find out!” I remember her being this great professor at MIT, and she was excited to find something that she didn’t know and go and learn about it together as opposed to being embarrassed that she didn’t know something. I learned a lot from that interaction: That it’s fun to not know something because then you get to go and find the answer, and no matter who you are, you’re never expected to know everything.

What is a PhD?

A PhD is a terminal degree, meaning it is the highest degree that you can earn. Almost without exception, people only earn one PhD. This degree is required for some jobs, mostly in academia, but is often just considered equivalent to years worked in industry. For some positions, hiring managers may even prefer to hire people without PhDs. A PhD is a paid (though not well paid) position at a university that lasts for between 4 and 10 years. To learn more about the PhD process, check out this previous post. 

The author on a field trip to Oceano Dunes in California. She is controlling RHex, a six-legged robot, outfitted with sensors to study dune migration.

The day-to-day life of a PhD student versus an industry professional

The process of earning the PhD is very different from the process of earning a bachelor’s or a master’s degree. It is more like an internship or a job. The first two or so years of any PhD program will be largely coursework, but even at this stage you will be balancing spending time on your courses against spending time on research – either because you are rotating through different labs, because you are performing research for a qualifier, or because your advisor is attaching you to an existing research project to give you some experience and mentorship before you develop your own project. This means that getting a PhD is not actually a way to avoid “getting a job” or to “stay in school” – it is actually a job. This also means that just because you are good at or enjoy coursework does not mean you will necessarily enjoy or excel in a PhD program, and just because you struggled with coursework does not mean you will not flourish in a PhD program. After you are done with coursework, you will spend all of your time on research. Depending on the day, that can mean reading textbooks and research papers, writing papers, making and giving presentations, teaching yourself new skills or concepts, programming or building robots, running experiments, and mentoring younger students. If you excelled at either conducting research as an undergraduate or very open-ended course projects much more than typical coursework, you’ll be much more likely to enjoy research as a PhD student. 

Types of goal setting in academia and industry

In course work, and in industry jobs with a good manager, you are given relatively small, well defined goals to accomplish. You have a team (your study group, your coworkers) who are all working towards the same goal and who you can ask for help. In a PhD program, you are largely responsible for narrowing down a big research question (like “How can we improve the performance of a self-driving car?”) into a question that you can answer over the course of a few years’ diligent work (“Can we use depth information to develop a new classification method for pedestrians?”). You define your own goals, often but not always with the advice of your advisor and your committee. Some projects might be team projects, but your PhD only has your name on it: You alone are responsible for this work. If this sounds exciting to you, great! But these are not good working conditions for everybody. If you do not work well under those conditions, know that you are no less brilliant, capable, or competent than someone who does. It just means that you might find a job in industry significantly more fulfilling than a job in academia. We tend to assume that getting a PhD is a mark of intelligence and perseverance, but that is often not the case — sometimes academia is just a bad match to someone’s goals and motivations. 

Meaning and impact

Academic research usually has a large potential impact, but little immediate impact. In contrast, industry jobs generally have an impact that you can immediately see, even if it is very small. It is worth considering how much having a visible impact matters to you and your motivation because this is a major source of PhD student burnout. To give a tangible example, let’s say that you choose to do research on bipedal robot locomotion. In the future, your work might contribute to prostheses that can help people who have lost legs walk again, or to humanoid robots that can help with elder care. Is it important to you that you can see these applications come to fruition? If so, you might be more fulfilled working at a company that builds robots directed towards those kinds of tasks instead of working on fundamental research that may never see application in the real world. The world will be better for your contributions regardless of where you make them – you just want to make sure you are going to make those impacts in a way that allows you to find them meaningful! 

Pay and lifetime earning potential

Engineers are significantly better paid in industry than academia. Since working in industry for a minimum of five to ten years and getting a PhD are often considered equivalent experience for the purposes of many job applications, even the time spent getting a PhD – where you will earn much less than you would in industry – can mean that you give up a substantial amount of money. Let’s say that an entry-level engineering job makes $100,000 per year, and a graduate student earns $40,000. If your PhD takes 6 years, you lose out on $60,000 x 6 = $360,000 of potential pay. Consider also that a PhD student’s stipend is fairly static, whereas you can expect to have incremental salary increases, bonuses, and promotions in an industry job, meaning that you actually lose out on at least $400,000. This is a totally valid reason to either skip the PhD process completely, or to work in industry for a few years and build up some savings before applying to PhD programs. 

Robotics Institute at University of Toronto

How do I know what I want?

It’s hard! If you’re still uncertain, remember that you can gain a few years of work experience in industry before going back to get the PhD, and will likely be considered an even stronger candidate than before. Doing this allows you to build up some savings and become more confident that you really do want to get that PhD. 

Thinking through these questions might help you figure out what direction you want to go:

• Are you much more motivated to do class projects that you are allowed to fully design yourself? 
• When you think about something small you built being used daily by a neighbor, how do you feel?
• Is your desire to get a PhD because of the prestige associated with the degree, or the specific job opportunities it opens up?

The basics

A PhD (doctoral degree) in engineering or a DEng (Doctorate of Engineering) is the highest degree that you can earn in engineering. This is generally a degree that people only earn one of, if they earn one at all. Unlike a bachelor’s degree or a master’s degree, a PhD studying a topic relevant to robotics should be free and students should receive a modest stipend for their living expenses. There are very few stand-alone robotics PhDs programs, so people generally join robotics labs through PhD programs in electrical engineering, computer science, or mechanical engineering. 

Joining a lab

In some programs, students are matched with a lab when they are accepted to the university. This matching is not random: If a university works this way, a professor has to have a space in their lab, see the application, and decide that the student would be a good fit for their lab. Essentially, the professor “hires” the student to join their lab. 

Other programs accept cohorts of students who take courses in the first few years and pick professors to work with by some deadline in the program. The mechanism through which students and professors pair up is usually rotations: Students perform a small research project in each of several labs and then join one of the labs they rotated in. 

The advisor

Regardless of how a student gets matched up with their advisor, the advisor has a lot of power to make their graduate school experience a positive one or a negative one. Someone who is a great advisor for one student may not be a great advisor for another. If you are choosing an advisor, it pays to pay attention to the culture in a lab, and whether you personally feel supported by that environment and the type of mentorship that your advisor offers. In almost every case, this is more important for your success in the PhD program than the specifics of the project you will work on or the prestige of the project, collaborators, or lab. 

Qualifiers

PhD programs typically have qualifiers at some point in the first three years. Some programs use a test-based qualifier system, either creating a specific qualifier test or using tests from final exams of required courses. In some programs, you are tested by a panel of faculty who ask the student questions about course material that they are expected to have learned by this point. In other programs, the student performs a research project and then presents it to a panel of faculty. 

Some universities view the qualifiers as a hurdle that almost all of the admitted PhD students should be able to pass, and some universities view them as a method to weed out students from the PhD program. If you are considering applying to PhD programs, it is worth paying attention to this cultural difference between programs, and not taking it too personally if you do not pass the qualifiers at a school that weeds out half of their students. After all, you were qualified enough to be accepted. It is also important to remember, if you join either kind of program, that if you do not pass your qualifiers, usually what happens is that you leave the program with a free master’s degree. Your time in the program will not be wasted!

The author testing a robot on a steep dune face on a research field trip at Oceano Dunes.

Research

Some advisors will start students on a research project as soon as they join the lab, typically by attaching them to an existing project so that they can get a little mentorship before starting their own project. Some advisors will wait until the student is finished with qualifiers. Either way, it is worth knowing that a PhD student’s relationship to their PhD project is likely different from any project they have ever been involved with before. 

For any other research project, there is another person – the advisor, an older graduate student, a post doc – who has designed the project or at least worked with the student to come up with parameters for success. The scope of previous research projects would typically be one semester or one summer, resulting in one or two papers at most. In contrast, a PhD student’s research project is expected to span multiple years (at least three), should result in multiple publications, and is designed primarily by the student. It is not just that the student has ownership over their project, but that they are responsible for it in a way that they have never been responsible for such a large open-ended project before. It is also their primary responsibility – not one project alongside many others. This can be overwhelming for a lot of students, which is why it is impolite to ask a PhD student how much longer they expect their PhD to take. 

The committee

The “committee” is a group of professors that work in a related area to the student’s. Their advisor is on the committee, but it must include other professors as well. Typically, students need to have a mix of professors from their school and at least one other institution. These professors provide ongoing mentorship on the thesis project. They are the primary audience for the thesis proposal and defense, and will ultimately decide what is sufficient for the student to do in order to graduate. If you are a student choosing your committee, keep in mind that you will benefit greatly from having supportive professors on your committee, just like you will benefit from having a supportive advisor. 

Proposing and defending the thesis

When students are expected to propose a thesis project varies widely by program. In some programs, students propose a topic as part of their qualifier process. In others, students have years after finishing their qualifiers to propose a topic – and can propose as little as a semester before they defend! 

The proposal and defense both typically take the form of a presentation followed by questions from the committee and the audience. In the proposal, the student outlines the project they plan to do, and presents criteria that they and their committee should agree on as the required minimum for them to do in order to graduate. The defense makes the case that the student has hit those requirements. 

After the student presents, the committee will ask them some questions, will confer, and then will either tell the student that they passed or failed. It is very uncommon for a PhD student to fail their defense, and it is generally considered a failure on the part of the advisor rather than the student if this happens, because the advisor shouldn’t have let the student present an unfinished thesis. After the defense, there may be some corrections to the written thesis document or even a few extra experiments, but typically the student does not need to present their thesis again in order to graduate.

The bottom line

A PhD is a long training process to teach students how to become independent researchers. Students will take classes and perform research, and will also likely teach or develop coursework. If this is something you’re thinking about, it’s important to learn about what you might be getting yourself into – and if it’s a journey one of your loved ones is starting on, you should know that it’s not just more school!

Photo credit: John Long.

Hypotheses about the evolution of traits in ancient species are difficult to test, as the relevant animals have often been extinct for thousands or millions of years. In the present study, a population of physical, free-swimming robots modeled after ancient fish evolved vertebrae under selection pressures for predator avoidance and foraging ability, showing how evolutionary robotics can be used to help biologists test hypotheses about extinct animals . 

Millions of years ago, during the Cambrian explosion, fish started to evolve tiny proto-vertebrae on the long flexible rods (notochords) that had previously given their bodies structure and lent some stiffness to their tails. As evolutionary traits go, vertebrae were very successful: they have been preserved through these millions of years, through fish, amphibians, reptiles, birds, mammals, and eventually ended up in your backbone.

So why did they evolve in the first place?

One theory is that the sudden increase of genetic diversity during the Cambrian explosion led to an “arms race” between predators and prey, the prize being either dinner (in the case of the predator) or your life (in the case of the prey). Take speed as an example: as prey animals evolve faster escape maneuvers to better evade a particular predator, the predatory species will be under more selection pressure to increase its speed as well. Arms races like this could have led to innovations like vertebrae, which enable fish to displace more water with every tail movement and thus swim faster for only a small increase in energy usage.

We were interested in testing how plausible it is that selection pressure for predator avoidance and foraging ability could drive the evolution of vertebrae, but unfortunately, all of the relevant animals were long extinct. Fortunately, when the desired study animals are unavailable, it’s often possible to create models of animal behavior using  other animals that have some characteristics in common with the species of interest, computer simulations of the animals and their environments, or physical simulations of the animals in similar environments in order to test hypotheses like those we were interested in here.

Computer simulations allow researchers to build huge populations of model organisms to which selection can be applied, but are not limited by the laws of physics. The behavior we were interested in here, which involves composite, flexible solids of varying stiffnesses bending in fluids, is difficult to accurately simulate with a computer; however, the number of generations that we wanted to be able to run to perform an evolutionary experiment would make building enough robots difficult. Rather than choose one method and be slave to its flaws, we did both: we created a physical simulation in our lab at Vassar and collaborated with two groups at Lafayette College to develop a computer simulation of the same system. If both of these simulations came up with similar results, there is stronger evidence that the results were not simply due to flaws in simulation.

At Vassar, we built a predator-prey system with one predator robot that did not evolve (Tadiator) and a population of six prey robots (Preyros) upon which selection was applied. Preyro was modeled a Devonian fish called Drepanaspis gemuendenensis, a jawless fish with an armored body, two eye-spots, and a short, flapping tail. This species was likely a bottom feeder, filtering out food particles from mud. From the fossil record (top image; Viennese Natural History Museum), artist Anton Fuerst created a 3D model of this fish (bottom left image; Viennese Natural History Museum), which we brought to (artificial) life in the autonomous, free-swimming robot Preyro (bottom right image).

Figure 1

Both predator and prey robots were controlled by an autonomous on-board microcontroller and propelled by a single servo motor that flapped a submerged tail. Preyro had two light sensors towards the front of its chassis, mimicking Drepanaspis’ eyespots, and two infrared emitter-receivers on its sides, mimicking a highly conserved motion sensor organ called the lateral line. Rather than eyespots and a lateral line, Tadiator had an IR receiver that was paired to an IR beacon on top of Preyro, which enabled Tadiator to constantly steer towards Preyro. The robots’ environment was a 10-foot diameter black tank in a darkened room whose only light source was a single lamp hanging close to the water.

Preyro had two behaviors, foraging and fleeing, which were controlled by a simple subsumption hierarchy. For foraging, Preyro would navigate towards the light source (its “food”) by looking at the difference in light between its left and right eyespots. Preyro was stimulated to flee whenever sensing either the wall or the Tadiator close to the IR emitter-receivers on its sides, and would initiate a “c-start” startle response. The c-start startle response is seen in all modern fishes, and very effectively moves the fish away from a potential predator. Preyro would then continue foraging as before. Tadiator had only  one behavior: seek Preyro.

Figure 2

We evolved a population of six Preyros over ten generations. Each individual Preyro was implemented on the same base Preyro hardware (same Handyboard, same motor, etc.) but with a swapped-out biomimetic tail and different code. Every individual Preyro in every generation was given three three-minute trials one-on-one with the Tadiator, where it had to forage for food while keeping safe from its predator. In a separate trial, we also measured peak acceleration during an escape attempt.

Figure 3

We applied two selection pressures to the population of Preyros: foraging ability, as measured by the average distance between Preyro and the center of the light source over the course of three three-minute trials; and predator avoidance, as measured by average speed during a trial, average distance from the predator robot peak acceleration during a startle response, and average number of successful escapes from the predator from three trials. A previous experiment indicated that selection for foraging ability alone could drive the evolution of stiffer backbones, but most of the variance of the data was still unexplained. Since this previous experiment indicated that foraging ability was important but not sufficient by itself, we included it along with predator avoidance as our two selection pressures.

Three character traits were allowed to evolve: the number of vertebrae, which controls the stiffness of the tail; the size of the caudal fin on the end of the tail; and the sensitivity of the lateral line. These three traits were chosen because they all appeared around the same time in evolutionary history, could be very important for foraging and fleeing, and vary greatly between species but are rarely lost completely.

Figure 4

Individuals in each generation were ranked according to a fitness function that rewarded successful escapes and successful foraging. The top three individuals from each generation were allowed to “mate” to produce the next generation, with the best-scoring individual contributing to six individuals in the next generation, the second best contributing to four, and the third best contributing to two. Genetic information from the parent generation was “mutated” by the addition of a little random noise and then randomly recombined. This gave the number of vertebrae the opportunity to also evolve by accident.

We measured the relative contributions of direct selection, chance, and indirect selection on each of these three character traits over each of the 10 generations. Allowing three traits to vary opened up the opportunity for the number of vertebrae to change as a by-product of trait correlation with some other trait, or for selection to act only on the caudal fin span and the predator detection threshold, indicating that number of vertebrae was not important for predator avoidance after all. An initial increase in the number of vertebrae was driven by direct selection alone, and as the average number of vertebrae in a generation leveled off at 5.7, chance and indirect selection became the primary drivers. In the last few generations, direct selection actually changed sign, indicating that the number of vertebrae had reached its local optimum.

We concluded that selection for predator avoidance along with foraging could indeed have driven the evolution of vertebrae in the Devonian period, using a combination of physical and computer models of organisms to build the next best thing to a time machine so that we could study animals that had been extinct for millions of years.

For more information, see the research paper:

Roberts SF, Hirokawa J, Rosenblum HG, Sakhtah H, Gutierrez AA, Porter ME and Long JH Jr (2014) Testing biological hypotheses with embodied robots: adaptations, accidents, and by-products in the evolution of vertebrates. Front. Robot. AI 1:12. doi: 10.3389/frobt.2014.00012

If you liked this article, you may also be interested in:

• Finding perfection in the imperfect: Applying Darwinian neuro-evolution to robotics
• ShanghAI Lectures: Josh Bongard “Morphological Change in Machines Accelerates the Evolution of Robust Behavior”
• Robots Podcast: Bio-inspired locomotion
• Robots Podcast: Curved Artificial Compound Eye

See all the latest robotics news on Robohub, or sign up for our weekly newsletter.