Abstract: Animals sustain the ability to operate after injury by creating qualitatively different compensatory behaviors. Although such robustness would be desirable in engineered systems, most machines fail in the face of unexpected damage. We describe a robot that can recover from such change autonomously, through continuous self-modeling. A four-legged machine uses actuation-sensation relationships to indirectly infer its own structure, and it then uses this self-model to generate forward locomotion. When a leg part is removed, it adapts the self-models, leading to the generation of alternative gaits. This concept may help develop more robust machines and shed light on self-modeling in animals.
Abstract: The way in which organisms create body schema, based on their interactions with the real world, is an unsolved problem in neuroscience. Similarly, in evolutionary robotics, a robot learns to behave in the real world either without re-course to an internal model (requiring at least hundreds of interactions), or a model is hand designed by the experimenter (requiring much prior knowledge about the robot and its environment). In this paper we present a method that allows a physical robot to automatically synthesize a body schema, using multimodal sensor data that it obtains through interaction with the real world. Furthermore, this synthesis can be either parametric (the experimenter provides an approximate model and the robot then refines the model) or topological: the robot synthesizes a predictive model of its own body plan using little prior knowledge. We show that this latter type of synthesis can occur when a physical quadrupedal robot performs only nine, 5-second interactions with its environment.
Abstract: This talk will outline challenges and opportunities in translating evolutionary learning of autonomous robotics from simulation to reality. It covers evolution and adaptation of both morphology and control, hybrid co-evolution of reality and simulation, handling noise and uncertainty, and morphological adaptation in hardware.
Abstract: Co-evolution of system models and system tests can be used for exploratory system identification of physical platforms. Here we demonstrate how the amount of physical testing can be reduced by managing the difficulty that a population of tests poses to a population of candidate models. If test difficulty is not managed, then disengagement between the two populations occurs: The difficulty of the evolved test data supplied to the model population may grow faster than the ability of the models to explain them. Here we use variance of model outputs for a given test as a predictor of the tests’ difficulty. Proper engagement of the co-evolving populations is achieved by evolving tests that induce a particular amount of variance. We demonstrate this claim by identifying nonlinear dynamical systems using nonlinear models and linear approximation models.
Abstract: Here we introduce one simulated and two physical three-dimensional stochastic modular robot systems, all capable of self-assembly and self-reconfiguration. We assume that individual units can only draw power when attached to the growing structure, and have no means of actuation. Instead they are subject to random motion induced by the surrounding medium when unattached. We present a simulation environment with a flexible scripting language that allows for parallel and serial self-assembly and self-reconfiguration processes. We explore factors that govern the rate of assembly and reconfiguration, and show that self-reconfiguration can be exploited to accelerate the assembly of a particular shape, as compared with static self-assembly. We then demonstrate the ability of two different physical three-dimensional stochastic modular robot systems to self-reconfigure in a fluid. The second physical implementation is only composed of technologies that could be scaled down to achieve stochastic self-assembly and self-reconfiguration at the microscale.
Abstract: Co-evolution is a biological process where populations of interacting individuals challenge each other in an ongoing cycle of adaptation, such as predator-prey competition and symbiotic cooperation. Though coevolution can potentially drive progress beyond stagnation point of conventional evolutionary algorithms, its application in practice has been challenging due to its complex dynamics. Here we show a systematic approach for implementing co-evolution in both the design and analysis of physical systems. The implementation is based on co-evolving predictive models and design steps. A number of applications in robotics and regulation networks are shown.
Abstract: Here we introduce a method for the evolution of dynamic gaits on a physical robot requiring no prior assumptions about the locomotion pattern beyond the fact that it should be rhythmic. The dynamic gaits were physically evolved in hardware using a parallel-actuated pneumatic robot. We have formu- lated a genetic algorithm that evolves open-loop controllers; the encoding allows evolution to shape both the speed and pattern of locomotion while ensuring rhyth- micity. In future, we plan to evolve closed-loop controllers for the physical robot and integrate our previously developed methods to reduce the number of hard- ware trials.