Since it’s large and this is unusual work, I’m dropping the whole abstract below. Also the preface. The two should give readers idea if it’s worth delving into the PDF. I only skimmed this a bit. It’s out of my area but did look neat. Thought some folks building physical things might find it interesting.
Abstract: “Evolutionary Design has been used to automatically generate a wide variety of novel and creative objects such as circuits, robots, and satellite antennae. And yet, despite the availability of sophisticated rapid prototyping machines capable of printing objects out of plastic, metal, and even circuitry, relatively few of these evolved designs have
been physically manufactured in the real world. We argue that the cause of this paucity of physical artifacts lies in the “design first, build later” philosophy of contemporary Evolutionary Design. By only specifying the form of an object, this approach leaves unanswered the vital question of formation. As evolved forms become more complex, their formation becomes increasingly difficult for both humans and computers to discover. As a consequence, there is a growing Fabrication Gap between the complexity of objects which we can evolve and those which we can manufacture.
The alternative proposed here is to use Artificial Ontogenies, a computational method inspired by the biological processes of growth, in order to directly evolve the formation of objects. We introduce Evolutionary Fabrication, the direct evolution of assembly instructions within a simulated manufacturing system, and show that this approach is capable of injecting the novelty and creativity associated with evolutionary approaches into the realm of fabrication, generating not just novel objects, but novel means of assembling those objects as well.
Ultimately, the evolution of form and formation become fully intertwined when the language of assembly itself becomes subject to evolution, capable of discovering increasingly large sub-assemblies and adding them to its vocabulary. Through this co-evolution of form and formation, Evolutionary Fabrication discovers both how to
build objects and what to build them out of. In this manner, Evolutionary Fabrication is capable of designing and assembling scalably complex objects in a hierarchical manner, even in the presence of error during assembly.
Via this co-evolution of form and formation, Evolutionary Fabrication circumvents the Fabrication Gap, leading the way to systems which can move from broad specification to complete artifact without the need for further human intervention. This budding field of Fully Automated Design and Manufacture will have an impact
on realms ranging from product design to planetary exploration.”
Preface: “When I first arrived at the DEMO Lab in the Fall 2001 I was given the task of assembling a variety of the “GenoBots” created by my colleague Greg Hornby. A few of the simpler robots were relatively easy to assemble, and produced some spectacular results - the most notable being the “QuadraBot” that walked in straight lines on four legs, each actuated by a single oscillating servo motor. Unfortunately, as the designs became more complex, the robots in turn became more difficult to build. The brittle plastic parts broke easily, and the actuators struggled to move ever larger masses. More frustrating, however, was the process of looking at one of Greg’s simulations and figuring out how exactly to build the shape on the screen. This was my first glimpse of the “Fabrication Gap”: the notion that there is a tremendous amount of missing knowledge between the blueprint of an object and the process required to physically manufacture it. And so, my motivation for this work stems from the same question that many
Ph.D. candidates ask themselves: “Can’t I get a computer to do this for me?” This, in the broadest sense, is the subject of this dissertation: how we can use Evolutionary Algorithms not just to design complex objects, but to assemble them as well.
Since it’s large and this is unusual work, I’m dropping the whole abstract below. Also the preface. The two should give readers idea if it’s worth delving into the PDF. I only skimmed this a bit. It’s out of my area but did look neat. Thought some folks building physical things might find it interesting.
Abstract: “Evolutionary Design has been used to automatically generate a wide variety of novel and creative objects such as circuits, robots, and satellite antennae. And yet, despite the availability of sophisticated rapid prototyping machines capable of printing objects out of plastic, metal, and even circuitry, relatively few of these evolved designs have been physically manufactured in the real world. We argue that the cause of this paucity of physical artifacts lies in the “design first, build later” philosophy of contemporary Evolutionary Design. By only specifying the form of an object, this approach leaves unanswered the vital question of formation. As evolved forms become more complex, their formation becomes increasingly difficult for both humans and computers to discover. As a consequence, there is a growing Fabrication Gap between the complexity of objects which we can evolve and those which we can manufacture.
The alternative proposed here is to use Artificial Ontogenies, a computational method inspired by the biological processes of growth, in order to directly evolve the formation of objects. We introduce Evolutionary Fabrication, the direct evolution of assembly instructions within a simulated manufacturing system, and show that this approach is capable of injecting the novelty and creativity associated with evolutionary approaches into the realm of fabrication, generating not just novel objects, but novel means of assembling those objects as well.
Ultimately, the evolution of form and formation become fully intertwined when the language of assembly itself becomes subject to evolution, capable of discovering increasingly large sub-assemblies and adding them to its vocabulary. Through this co-evolution of form and formation, Evolutionary Fabrication discovers both how to build objects and what to build them out of. In this manner, Evolutionary Fabrication is capable of designing and assembling scalably complex objects in a hierarchical manner, even in the presence of error during assembly. Via this co-evolution of form and formation, Evolutionary Fabrication circumvents the Fabrication Gap, leading the way to systems which can move from broad specification to complete artifact without the need for further human intervention. This budding field of Fully Automated Design and Manufacture will have an impact on realms ranging from product design to planetary exploration.”
Preface: “When I first arrived at the DEMO Lab in the Fall 2001 I was given the task of assembling a variety of the “GenoBots” created by my colleague Greg Hornby. A few of the simpler robots were relatively easy to assemble, and produced some spectacular results - the most notable being the “QuadraBot” that walked in straight lines on four legs, each actuated by a single oscillating servo motor. Unfortunately, as the designs became more complex, the robots in turn became more difficult to build. The brittle plastic parts broke easily, and the actuators struggled to move ever larger masses. More frustrating, however, was the process of looking at one of Greg’s simulations and figuring out how exactly to build the shape on the screen. This was my first glimpse of the “Fabrication Gap”: the notion that there is a tremendous amount of missing knowledge between the blueprint of an object and the process required to physically manufacture it. And so, my motivation for this work stems from the same question that many Ph.D. candidates ask themselves: “Can’t I get a computer to do this for me?” This, in the broadest sense, is the subject of this dissertation: how we can use Evolutionary Algorithms not just to design complex objects, but to assemble them as well.