Justin Werfel

Designing collective behavior in a termite-inspired robot construction team.

Robots programmed with simple construction rules can work independently but collectively to build a complex structure. [Also see Perspective by Korb] Complex systems are characterized by many independent components whose low-level actions produce collective high-level results. Predicting high-level results given low-level rules is a key open challenge; the inverse problem, finding low-level rules that give specific outcomes, is in general still less understood. We present a multi-agent construction system inspired by mound-building termites, solving such an inverse problem. A user specifies a desired structure, and the system automatically generates low-level rules for independent climbing robots that guarantee production of that structure. Robots use only local sensing and coordinate their activity via the shared environment. We demonstrate the approach via a physical realization with three autonomous climbing robots limited to onboard sensing. This work advances the aim of engineering complex systems that achieve specific human-designed goals. Robot Rules In the case of mound-building termites, colonies comprising thousands of independently behaving insects build intricate structures, orders of magnitude larger than themselves, using indirect communication methods. In this process, known as stigmergy, local cues in the structure itself help to direct the workers. Werfel et al. (p. 754; see the Perspective by Korb) wanted to construct complex predetermined structures using autonomous robots. A successful system was designed so that for a given final structure, the robots followed basic rules or “structpaths” in order to complete the task.

BCI competition 2003-data set Ia: combining gamma-band power with slow cortical potentials to improve single-trial classification of electroencephalographic signals

In one type of brain-computer interface (BCI), users self-modulate brain activity as detected by electroencephalography (EEG). To infer user intent, EEG signals are classified by algorithms which typically use only one of the several types of information available in these signals. One such BCI uses slow cortical potential (SCP) measures to classify single trials. We complemented these measures with estimates of high-frequency (gamma-band) activity, which has been associated with attentional and intentional states. Using a simple linear classifier, we obtained significantly greater classification accuracy using both types of information from the same recording epochs compared to using SCPs alone.

Three-Dimensional Construction with Mobile Robots and Modular Blocks

We present a decentralized algorithmic approach to automatically building user-specified three-dimensional structures from modular units. Our bipartite system comprises passive units (blocks), responsible for embodying the structure and determining where further units can legally be attached, and active units (robots), responsible for transporting passive units. The algorithmic issues are correspondingly decomposed into two parts: (1) deciding where passive units may be attached; and (2) getting them to those locations. For the first part, we give simple, scalable rules for attachment and prove that they will reliably lead to the construction of any desired structure from a large class of three-dimensional shapes. For the second part, we compare three approaches: random movement, systematic search and gradient-following; each approach is successively faster but requires more communication overhead and/or unit capabilities. The system we describe enables guaranteed construction of desired structures using very simple agent algorithms, taking a high-level specification as the only required input. The topic of collective construction is related to the problems of programmed self-assembly and self-reconfiguration in modular robots, and the rules governing block attachment presented here may be usefully applied to such systems.

Distributed construction by mobile robots with enhanced building blocks

We describe a system in which autonomous robots assemble two-dimensional structures out of square building blocks. A fixed set of local control rules is sufficient for a group of robots to collectively build arbitrary solid structures. We present and compare four versions in which blocks are (1) inert and indistinguishable, (2) uniquely labeled, (3) able to be relabeled by robots, (4) capable of some computation and local communication. Added block capabilities increase the availability of nonlocal structural knowledge, thereby increasing robustness and significantly speeding construction. In this way we extend the principle of stigmergy (storing information in the environment) used by social insects, by increasing the capabilities of the blocks that represent that environmental information. Finally, we describe hardware experiments using a prototype capable of building arbitrary solid 2-D structures

TERMES: An Autonomous Robotic System for Three-Dimensional Collective Construction

Collective construction is the research area in which autonomous multi-robot systems build structures according to user specifications. Here we present a hardware system and high-level control scheme for autonomous construction of 3D structures under conditions of gravity. The hardware comprises a mobile robot and specialized passive blocks; the robot is able to manipulate blocks to build desired structures, and can maneuver on these structures as well as in unstructured environments. We describe and evaluate the robot’s key capabilities of climbing, navigation, and manipulation, and demonstrate its ability to perform complex tasks that combine these capabilities by having it autonomously build a ten-block staircase taller than itself. In addition, we outline a simple decentralized control algorithm by which multiple simultaneously active robots could autonomously build user-specified structures, working from a high-level description as input.

Learning Curves for Stochastic Gradient Descent in Linear Feedforward Networks

Gradient-following learning methods can encounter problems of implementation in many applications, and stochastic variants are sometimes used to overcome these difficulties. We analyze three online training methods used with a linear perceptron: direct gradient descent, node perturbation, and weight perturbation. Learning speed is defined as the rate of exponential decay in the learning curves. When the scalar parameter that controls the size of weight updates is chosen to maximize learning speed, node perturbation is slower than direct gradient descent by a factor equal to the number of output units; weight perturbation is slower still by an additional factor equal to the number of input units. Parallel perturbation allows faster learning than sequential perturbation, by a factor that does not depend on network size. We also characterize how uncertainty in quantities used in the stochastic updates affects the learning curves. This study suggests that in practice, weight perturbation may be slow for large networks, and node perturbation can have performance comparable to that of direct gradient descent when there are few output units. However, these statements depend on the specifics of the learning problem, such as the input distribution and the target function, and are not universally applicable.

Massive uniform manipulation: Controlling large populations of simple robots with a common input signal

Roboticists, biologists, and chemists are now producing large populations of simple robots, but controlling large populations of robots with limited capabilities is difficult, due to communication and onboard-computation constraints. Direct human control of large populations seems even more challenging. In this paper we investigate control of mobile robots that move in a 2D workspace using three different system models. We focus on a model that uses broadcast control inputs specified in the global reference frame. In an obstacle-free workspace this system model is uncontrollable because it has only two controllable degrees of freedom - all robots receive the same inputs and move uniformly. We prove that adding a single obstacle can make the system controllable, for any number of robots. We provide a position control algorithm, and demonstrate through extensive testing with human subjects that many manipulation tasks can be reliably completed, even by novice users, under this system model, with performance benefits compared to the alternate models. We compare the sensing, computation, communication, time, and bandwidth costs for all three system models. Results are validated with extensive simulations and hardware experiments using over 100 robots.

Multilevel and Kin Selection in a Connected World

Arising from: G. Wild, A. Gardner & S. A. West 459, 983–986 (2009)10.1038/nature08071; Wild, Gardner & West replyWild et al. argue that the evolution of reduced virulence can be understood from the perspective of inclusive fitness, obviating the need to evoke group selection as a contributing causal factor. Although they acknowledge the mathematical equivalence of the inclusive fitness and multilevel selection approaches, they conclude that reduced virulence can be viewed entirely as an individual-level adaptation by the parasite. Here we show that their model is a well-known special case of the more general theory of multilevel selection, and that the cause of reduced virulence resides in the opposition of two processes: within-group and among-group selection. This distinction is important in light of the current controversy among evolutionary biologists in which some continue to affirm that natural selection centres only and always at the level of the individual organism or gene, despite mathematical demonstrations that evolutionary dynamics must be described by selection at various levels in the hierarchy of biological organization.

Building Blocks for Multi-robot Construction

One notable capability of social insect colonies that has traditionally inspired distributed robot systems is their construction activity. In this paper, I describe a system of simple, identical, autonomous robots able to build two-dimensional structures of arbitrary design by rearranging blocks of building material into desired shapes. Structure design is specified compactly as a high-level geometric program; robots translate this program into physical form via their fixed behavioral programming. Robots are interchangeable both within and between construction projects, and need not be individually reprogrammed between dissimilar projects. Such a construction team could be used as the first stage in a system for remote building of structures, laying out the floor plan that a more sophisticated system could extend upwards.

How Changes in Extracellular Matrix Mechanics and Gene Expression Variability Might Combine to Drive Cancer Progression

Changes in extracellular matrix (ECM) structure or mechanics can actively drive cancer progression; however, the underlying mechanism remains unknown. Here we explore whether this process could be mediated by changes in cell shape that lead to increases in genetic noise, given that both factors have been independently shown to alter gene expression and induce cell fate switching. We do this using a computer simulation model that explores the impact of physical changes in the tissue microenvironment under conditions in which physical deformation of cells increases gene expression variability among genetically identical cells. The model reveals that cancerous tissue growth can be driven by physical changes in the microenvironment: when increases in cell shape variability due to growth-dependent increases in cell packing density enhance gene expression variation, heterogeneous autonomous growth and further structural disorganization can result, thereby driving cancer progression via positive feedback. The model parameters that led to this prediction are consistent with experimental measurements of mammary tissues that spontaneously undergo cancer progression in transgenic C3(1)-SV40Tag female mice, which exhibit enhanced stiffness of mammary ducts, as well as progressive increases in variability of cell-cell relations and associated cell shape changes. These results demonstrate the potential for physical changes in the tissue microenvironment (e.g., altered ECM mechanics) to induce a cancerous phenotype or accelerate cancer progression in a clonal population through local changes in cell geometry and increased phenotypic variability, even in the absence of gene mutation.