“Dual Pneuma was initiated as this balancing, state-changing machine, and it evolved into a tool to make ceramics. As the ideas get more complex, they begin to reference the past, the future, the quality of being alien or ancient.”

David Familian: There are multiple areas of feedback in your work. Describe how your work uses feedback, and then, in general, how your works have dealt with the issue of feedback in various ways and how, in particular, you went from hydraulic to pneumatic muscles.

Chico MacMurtrie: Feedback is an integral element of a robotic system. I haven’t necessarily always worked strictly in robotics. I’ve used machines that didn’t have feedback. Several of these machines needed encoders or sensors—early mechanical works before hydraulics and pneumatics were introduced. For example, Totemobile, a work from 2007 commissioned by Citroën, used hard pneumatic systems with actuators, cylinders, electric motors, and a super-complex feedback system of sensors, proximity sensors, encoders, and read switches for positioning. So, there were over 350 different sensors on that system. There are 24 separate machines in that big machine. It’s run with a ladder logic and a redundant sensing system that looks at every sequence to ensure that it runs safely. And it’s completely dependent on feedback. The robotic skeletons of Skeletal Reflections had onboard encoders on it. But the interaction came from the viewer striking a pose, which was read by an early vision system and would then be turned into a little line drawing. The machine would then search the library for that pose, and then the machine would perform that pose. So it would interpolate from one posture to the next. 

The inflatables that we’re working on now aren’t new to feedback, because what we used for most of the inflatable architecture work was what we call pressure sensors, where we would measure the pressure in the machine. Then, we used that pressure on the machine to help close the loop and determine how we got the machines into different positions. Dual Pneuma, the work in the show, uses something even simpler. We’re using bend sensors which give a different value when you bend it. And that’s how you’re able to get the machine into different positions. What we have is the comparison between how artists work on robotics versus, say, a roboticist or a research student. The scientists are working on it through learning from past systems, and responding to papers that have been already addressed on the subject, and they just keep elaborating on that. So they stick real close to the lessons of traditional robotics. All these walking machines you see online are still hard-linked machines driven by motors and actuators.

“Scientists stick real close to the lessons of traditional robotics. All these walking machines you see online are still hard-linked machines driven by motors and actuators.”

When we were short on time, I told the PhD students to move on from the forward-and-reverse kinematics studies, stop taking the robot apart to try to get it to draw a C, get the robot on all four feet, have it respond to gravity, and start teaching it how to walk. So that’s what we’re doing over here at the lab. The approach we’re considering is to drive it into positions and then save those positions as presets. Now we have the option of actually recording actual movements, using a small puppet which I can use directly to drive the machine. That’s a series of encoders on a little apparatus that you can puppet with your fingers, and the machine responds directly. And then it records those motions.

Another layer is that the four points of contact will have these switches on them to know when the four points are touching the ground or if it’s three points. The way you drive it to navigate the space is that you set a rule in the code so that you always have three points of contact. So, the feedback system is the bend sensors at the joints and the pressure sensors at the feet. That’s the approach we’re using at the moment. Probably what will happen is we’ll have the bigger one do more navigation type of stuff, and the smaller one would probably do more organic things on the ground, like perform Yoga-types of positions and things of that nature.

David Familian: And how do you think the piece builds from the feedback to become more complex? 

Chico MacMurtrie: What the team at UC San Diego is trying to do is to use math and a range of references in traditional robotics, but they’re having a hard time leaning into the organic nature of Dual Pneuma. Their data sets are generated from hard machines, whereas this is a soft machine. There are two layers of feedback—one from the bend sensors and one from the pressure sensors. Part of what we’re doing now is we’re having to do some graph calculations for the resolution of the encoders.

David Familian: Maybe that has to be articulated in some way in the didactic, that in working with traditional roboticists, it became unwieldy to come up with the calculations, and it was easier for you to make your own puppet and use your mind as the processor to come up with ideas.

Chico MacMurtrie: I think what would be interesting to include is the variable techniques we went through to try to get these things working. What Allyson Chen is coding with mayy lab and they’re gonna set up a way to run different behaviors from that little machine with sensors or something.

David Familian: You’re taking the code you’ve developed with your machine and building from that, instead of starting from scratch?

Chico MacMurtrie: No, we’re starting from what Allyson created for our show at UCSD. She can get it to do things just by writing code and seeing what happens. And if necessary, we can hook her up with a little machine like what we’re working with. But the machine we’re working with will be helpful in terms of our ability to program the complexity of this new piece.

David Familian: I think it’s interesting to work with the totality of the physical remote model, as opposed to looking at a screen and moving a virtual model around. It seems like for you and the way you work, it’s better for you to have this than a virtual model on the screen.

Chico MacMurtrie: That’s right. Even if you really command that medium of a virtual model, you’re still up against real-life situations, right? So when you’re running robots in the real world, and a human performer is interfacing to drive it to the left and the right. No matter how good the model and video are, they will be insufficient at meeting the complexity of live scenarios. We’re using techniques similar to theirs. They just have a lot more sophisticated software for balance and stuff.

“Even if you really command a virtual model, you’re still up against real life. No matter how good the model is, it will be insufficient at meeting the complexity of live scenarios.”

David Familian: I’m intrigued by the fact that out of the frustration of not getting exactly what you wanted, you were forced to come up with another solution, which you may not have previously considered. Did you have a feeling you’ll have to do this anyway, at some point?

Chico MacMurtrie: Yeah, I figured that I would have returned to a previous method of robot control that I had used in the past.

David Familian: Like the control pad, or something similar, where you turn something and watch as the robot mirrors your commands. I remember you kept saying that you could control it with the usual interface. And you got it to happen. And I think that’s something to talk about a little bit in the video, too, is the challenges of this work which were not present in previous projects that didn’t need a living model, I mean a scaled-down model to work. You could intuitively work those big robots with just a few dials,  but this is a more complex system, with reverse kinematics and all that stuff going on. So it’s tough to figure that out when you’re turning knobs.

Chico MacMurtrie:  Also, with your standard slider on a Max patch, we could have 10 or 20 sliders, and you can only really move one at a time. With that robot, you could record one movement at a time on the arm and just turn that into a subsequence and keep doing that, or you could try to re-record a whole action with it. We’ll likely end up using a combination of both the new and old methods to control this robot. The truth is that we built a tool to allow me to control it and get it to do what I want for two reasons: one, it’s tough to tell the programmer what I want in animation. It takes a long time, but you can finesse the subtlety of it.

David Familian: Allyson was doing what you’re doing. She was programming it step by step herself. Her procedure is a lot harder because she has to do it a little bit at a time, and it’s a lot more trial and error, whereas yours limit the trial and error that she had to fight with. Each time she made a movement to make sure everything else followed correctly. But your model follows where you want to move it.

Chico MacMurtrie: That’s right, and we probably won’t get an extremely precise following, but you can teach it behaviors by looking at it and moving it and getting it to do something, so you’ll be able to control your puppet and get a feel for it directly. It’s exactly the way the Border Crossers (2021) had to work. With that project, we abandoned turning it into a machine completely because we had different terrain and wind conditions to deal with. So, we ended up building an interface that was more suited for a human driver. The difference here is we’re going to use that device to record what we’ll end up using for the show, and that’s in sequences. Right? So you build up a series of sequences, and that sequence is just ‘move forward, turn left, bow down,’ etc.

David Familian: So, will your presentation at the Beall be able to enact these different recorded keyframes, or will it pick from a random set of movements from time to time?

Chico MacMurtrie: I got the folks at UCSD working on a sensing system to use the sensing system to gather data on audience movements so that if people trigger a certain sensor, certain behaviors will be enacted. We’ll figure out how to trigger the more complicated things, for when you have an audience that should see the complicated stuff, and then some simpler things for the everyday audience.

David Familian:  Do you think when you’re installing, could our choreographer Kelli Sharp work with that model and translate its movements to the robot? Or is it a lot of work to encode movements.

Chico MacMurtrie: If Kelli could use the facility that she has, it would be more ideal for her to use a motion capture system that she could wear, which would be taking the role of the puppet.

David Familian: Right, I just thought it might require less translation.

Chico MacMurtrie: The idea behind the ceramics and the physical inflatables is that the inflatables pay homage to the ancestors. They are the offspring of the ceramics. And they’ll be making different kinds of sounds that set the mood of the installation.

David Familian: Can you articulate in some way how working on this piece has brought this idea of complex systems more to the fore of your practice.

Chico MacMurtrie: Sure. At the beginning of the project, I was thinking about doing more complex types of actions, utilizing balance and stuff like that. And I hadn’t really taken that on, and I’ve only recently begun to touch on the foot sensors. Early on, I was considering this humanoid form, and it took me to another place of thinking. At the point when you approached me, it triggered the beginning of this piece that had been in my mind for many years, but I didn’t have the means or the right opportunity to do it. Results were great in the beginning because there was some funding to do it, and then it also triggered the other works that are going to be in the show.

Dual Pneuma was initiated with the notion of this really complex balancing, state-changing machine, and it evolved into a tool to make ceramics. And so as the ideas get more complex, they begin to reference the past, the future, and even the quality of being alien or the product of ancient cultures. Dual Pneuma evokes a feeling that differs from the feelings you see in the more polished robots you encounter online, whose movements are precise and premeditated. In this project, viewers see how the robots move, and viewers have no sense of what they’re going to do, which makes them unique.

“The machine transitions between states of crawling and walking. When it seems to be falling over, it transitions into a different posture. It has this morphing ability to go from one state to the next.”

David Familian: And the materiality makes them more organic! I’m also curious why this piece pushed you to make fixed sculptures, which you’d never made before. Also, do you think this work is more alive than your other works?

Chico MacMurtrie: Well, it has infinite arrangements that it can do. One of the simple exercises that we got it to do on video yesterday was to use the machine as a plotter and to plot shapes, draw letters and stuff. He’s trying to figure out how to do this stuff. And so he’s going through all these exercises with this contraption you would normally use like a XYZ thing for right. And he realized that we have 350 different possibilities here. 

You have the action of an animal, an insect, an amoeba, a cell, a germ—it takes on all the qualities of all the things I’ve explored in 35 years of work because it’s boneless, in a sense. Its joints flex universally. While humans can imitate animals in an interesting way, it takes an athletic and dexterous human to do that. It can be multiple things, and that in itself is quite complex. As you move towards a deadline of an exhibition or a presentation, you have endless possibilities of what it can do, but you have to choose what you’re going to have it do to keep it cohesive and intact.

David Familian: My final question is: When we initially talked, I wanted you to make the viewer viscerally feel the relationship between order and disorder. How do you think the work is accomplishing that now? Right now, there’s a certain kind of awkwardness in the work because it’s like a baby learning to walk. It’s a little more advanced but it moves weirdly and doesn’t seem like a traditional animal because it’s learning. And you’re teaching it as you would train an animal. In your work, I see the tension between order and disorder in this space between smooth movement and awkwardness when the robot misses its mark.

Chico MacMurtrie: Yeah, I think you’re right about that. The smaller machine went between states of crawling, walking, and falling over. It has this morphing ability to go from one thing to the next. So, instead of being an accident or a fall, it can merge into another position. We’re concentrating on getting it to do some of these complex things I set out to do. The larger one will be more focused on walking around and doing things more uprightly. And if it falls over, it’ll fall into a seated position.

→ Chico MacMurtrie, Dual Pneuma, 2024