Humans are made of cells. Robots are made of parts. Our ability to sense and react to our environments, to truly feel both the things we touch and our own physical movements is part of what separates us from machines.
But in the world as Ben Goertzel envisions it, robots likewise could be comprised of cells – small spherical components that serve as both sensors and actuators, just as the cells in our bodies do. These macrocells, as Goertzel calls them, would solve what he considers to be the biggest limitation of current robots: the lack of kinesthesia, or the ability to feel the parts of your body as they act. Combined to form an inorganic macrocell robot, or Imbot, the components would need a way to connect to and communicate with each other, so the robot as a whole could react to stimuli, even changing its shape to navigate its surroundings.
Geortzel is no roboticist, and he readily admits it. But he comes to this problem with a considerable knowledge base and an impressive pedigree. Goertzel developed the open source artificial general intelligence software OpenCog and serves as CEO of two technology companies: Biomind, a bioinformatics company that applies AI to genetics research, and Novamente, an AI consulting company. After working with several commercial robots and growing frustrated with the kind of sensory data they were able to provide, he became convinced that the way to solve the AI problems he was facing was by building a radically different kind of bot.
Goertzel described his idea for the Imbots at length in a piece for H+ magazine (published by the nonprofit organization Humanity+, of which Goertzel is Vice Chairman) in August, describing how the macrocells could be specialized, much like human cells are, to serve as eyes, ears, speakers, power sources or muscles. In addition to needing to contain a programmable microprocessor, be able to connect to other cells and sense pressure, the muscle cells would also have the ability to deform into different shapes, based on instructions it receives in response to pressures in the environment.
Modular robotics itself is nothing new. Anyone who has seen Transformers is at least somewhat familiar with the idea of shape-shifting robots. The key difference with an Imbot is the sensory aspect. One macrocell senses something, say an obstacle, and passes the information to all the other cells. The flexible, marble-sized balls would work together to navigate the challenge.
Imagine a Roomba-type vacuum cleaner bot made in this way, Geortzel says. “If it had to crawl across the mess on the floor in my house it could just deform and change its shape to crawl over the mess because every part of it is a sensor. It would be able to react more like a biological organism.”
Nanoscale Bots at CMU
The Imbots, for now, are just an idea. But there is similar, and perhaps even more ambitious, research being done at the Claytronics lab at Carnegie Mellon University. Claytronics is, in many ways, a nanoscale version of Goertzel’s macrocell bots. The project’s researchers are focusing on building miniature robots that they call catoms (for Claytronics atoms) that can then arrange themselves into any configuration as instructed by a computer program. With millions upon millions of catoms, you would get completely programmable matter. As opposed to Goertzel’s marble-sized macrocells, Claytronics is shooting for particles on the order of a couple hundred microns across.
Many of the same challenges that would theoretically apply to building Imbot macrocells, also apply to creating catoms: you need a wireless power source, surface-to-surface communication, and the ability to run controls. The catoms will be made through a MEMS process – microelectricalmechanics, which is essentially the same process used to make computer chips. “Instead of making a flat surface, we’re engineering a process that prints the computer chip flat, then has it fold up into a sphere,” explains Seth Goldstein, director of the Claytronics lab and a professor at CMU. One of Goldstein’s students is on the verge of demonstrating such a device – a millimeter-wide cylinder that meets all those qualifications. As a project that is still in its early stages, Goldstein explains that large amounts of the Claytronics programmable matter – enough to form a car, for example, are still around 10 years away.
“If they succeed, that would be amazing,” Goertzel says. “I think they will succeed, eventually.” His Imbot idea is meant to be something a little simpler, and shorter-term than Claytronics that will still provide artificial intelligence researchers with a more flexible robot infrastructure. While Claytronics is focused on making completely programmable matter that can form itself into any shape imaginable provided there are enough catoms to make it, Goertzel’s Imbots are less concerned with complete reconfigurability and more concerned with flexibility and the integration of movement with sensation. He’s interested in humanoid bots, and human-level AI systems and, as he puts it, “I couldn’t grow a third arm, even if I was willing to reuse the bones in my leg to make that third arm. There are limitations to how reconfigurable a human body is.”
Goertzel says he doesn’t believe Imbots are fundamentally any more difficult to build than robots built by current methods. Of course, building something like this would essentially be like starting from scratch, and there would be a learning curve that would necessarily make the Imbots less precise, at least for a time. Not something that should be implemented for, say, surgical bots, where accuracy is key. But for a service robot, the ability to adapt to different environments and the ability to feel its body (being aware, for example, of how much pressure it was putting on a human) might be more important than precision.
This initial lack of precision is a result of the mathematics behind the programming. For conventional robots with a limited number of joints, solving the equation to get them to work in tandem is far less difficult than it would be for a bot composed of hundreds or thousands of macrocells. The math problem becomes harder, and the learning curve grows steeper as you add more flexible parts. “On the other hand,” Goertzel says, “potentially you could do the movements better once it [the Imbot] has learned.”
Less Than 100 Lines of Code
While these kind of robots would be more difficult to program, it is by no means an impossible task. The Claytronics project is developing new programming languages and “new programming paradigms, which are actually pretty successful for doing things like shape change,” Goldstein says. On YouTube, there’s a 30-second video that shows a model of a 10-million catom square morphing into the new Gates Hillman Complex, a building at the CMU computer science department. What is remarkable about this video is that, not only is the program being executed using less than 100 lines of code, it’s actually provably correct. That is, the CMU team has proven that the algorithm will get to the final shape in a finite amount of time, without getting stuck and without wasted effort. This, Goldstein says, is a big deal in the programming community.
In the end, though the similarities between Claytronics and Goertzel’s Imbot idea may prove useful for programmers, the differences in scale and focus mean they will serve very different purposes. Goldstein says that artificial intelligence is orthogonal to the Claytronics Lab’s research, but imagines AI researchers would probably be interested in it. And indeed, Goertzel says his company, Novamente, would want to be involved should something like Claytronics or the Imbots ever be built. Since the macrocell bots would both enable and require more advanced AI systems, the two would have to develop alongside one another.
Clearly, nothing like this exists yet on a large scale beyond the realm of imagination. But since Goertzel’s article was published in August, he has received some responses, mostly from “young guys from the DIY hardware community,” he says. However, he has also talked with Samsung’s humanoid robotics lab, and they’re interested in the concept. Whether or not anything will come of it, Goertzel says, is hard to tell, “but the conversation is ongoing.”
For his part, Goldstein suggests thinking of Claytronics as a new media type – something that will completely change our idea of what robotics is, or can be. You might swallow a clump of claytronics at the hospital that changes shape with your intestines to capture images, or stop completely and perform surgery. Goertzel’s Imbots, even if they get built, won’t be doing anything quite that cutting-edge, but should developers get behind the idea, flexible, kinesthetic macrocell robots could be a stepping stone toward this radical new media, and one that addresses many of robotics’ current problems in the process.