Evolutionary Robotic Prototype
Supported through the technological advances and the relative fields of expertise, robotics, computation and engineering; architecture is considered to be at a transition stage engaged with time and mobility. Kinetic architecture aims at the development of timely dynamically adaptable systems differing to external or internal conditions, leading to buildings or building components with variable mobility or geometry.
Recently, significant work in the area of adaptive structures and mobile robotics incorporates methods of anticipating unexpected situations that can be categorised into two section based on their error diagnosis methodology. Specifically examples are categorised into the ones using online operation by continuous tests on the physical robot (Srinivas, 1977; Mahdavi & Bentley, 2003), and the others using offline processes for which the trials rely mostly on simulated robot minimising the number of hardware trials applied (Baydar & Saitou, 2001; Lipson et al., 2005; 2006). Both categories emphasise in the ability of a robot to function for long periods of time in an unknown, remote environment, by being able to deal with uncertainty without any human operators to manually repair or provide compensation to changes occurred. The central question hereby is how structures could still be considered useful even when they experience damage. Within this context, a physical and digital prototype is investigated, capable of mobility and locomotion. By extending current researches, the hypothesis is that an offline method would be most efficient to gain an understanding of current situation at each time step.
The objective of such robots is to maintain and improve their usability and performance by recognising changes in their morphology and providing new behaviours accordingly. The goal of this research follows the direction of the offline error diagnosis methodology investigated by Hod Lipson et al. (2005; 2006) as an approach of maintaining a morphological understanding of the robot by optimising the connections between the nodes defined. The methodology follows a co-evolutionary approach of the estimation-exploration algorithm applied onto a robotic framework. The algorithm optimises the actions to be applied onto the robot by finding the ones that reveal most information about its morphology, while simultaneously co-evolving the simulated models of the target robot until minimum error between the models and the robot is reached. By enabling a robot to maintain an â€˜imageâ€™ of itself will provide the ability of the robot to understand when a difference occurred and therefore behave meaningfully in the newly defined context.
Mobile robotics face the possibility of damage under the process of locomotion into rough terrains. Furthermore, the project investigates locomotion strategies into rough, unstructured terrain of a robot after obtaining an overall morphological understanding of itself. The control system is based on evolutionary computation processes (Agogino et al., 2013; Iscen et al., 2013; Mirletz et al., 2015), emphasising in adaptive strategies of evolving behaviour that is not predefined, but rather is investigated. Locomotion ability is considered a non-deterministic task that provides little information relative to the overall optimised behaviour under investigation, having no clear connection between input and output. Therefore, instead of analogically defining a control policy into the robot, evolutionary algorithms are used to generate locomotion pattern of the prototype. The optimised behaviour emerges through iterations of evaluations of the resulted pattern, improving the performance in each time step. The gait investigated follows a â€˜flop and rollâ€™ method (Iscen et al., 2014) and the suggested robot is defined accordingly.
Within this frame of consideration, tensegrity structures are chosen as compliant, lightweight structures of nonlinear dynamics that have the ability of deploying and kinematically transforming through the active control of their network of members. The structure investigated is defined as a tensegrity system with the modification of a fixed centroid mass where the activation mechanisms are attached. The prototype suggested, consists of 4 compression members, stabilized through 18 tensile pretensioned members. Each strut linkage is a motion actuator of fixed length performing locally a sliding movement in both directions (up and down), thus shifting the overall member, connected with passive elastic linkages; providing shape deformation to the overall structure. The kinematic transformation is achieved through the linear control of the strut members and the respective relative length deformation of the adjacent tensile members. Sensor readings rely onto a 9DOF accelerometer located in the centroid mass, providing information about the prototypeâ€™s orientation, while continuously informing the digital model. Information between the sensors readings and actuatorsâ€™ values defines the exact situation of the robot, while any change in the initial configuration (target robot), suggest that damage has been occurred and the robot needs to maintain a new morphological understanding.
Agogino, A. SunSpiral, V. & Atkinson, D. (2013). Super Ball Bot â€“ Structures for Planetary Landing and Exploration. NASA Innovative Advanced Concepts (NIAC) Program, Final Report.
Baydar , C.M. & Saitou, K. (2001). Off-line error prediction, diagnosis and recovery using virtual assembly systems. In IEEE Intl. Conf. on Robotics and Automation, 818â€“823.
Glynn, R. & Bondin, W. (2014). Fabricating Behaviour: a problem, a solution, a prototype and a proposal, Fabricate: Negotiating Design and Making, gta Verlag.
Iscen, A., Agogino, A., SunSpiral, V., & Tumer, K. (2013). Controlling Tensegrity Robots through Evolution,Â In Proceedings of Genetic and Evolutionary Computation Conference, (GECCO 2013), Amsterdam, The Netherlands, July 6-10, 2013
Iscen, A., Agogino, A., SunSpiral, V., Tumer, K. (2014). Flop and Roll: Learning Robust Goal-Directed Locomotion for a Tensegrity Robot, In Proceedings of The 2014 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS 2014), Chicago, USA.
Lipson H., Bongard J., Zykov V. (2006). Resilient Machines Through Continuous Self-Modeling, Science, 314(5802), 1118 â€“ 1121.
Lipson H., Bongard J. (2005). Automatic Synthesis of Multiple Internal Models Through Active Exploration, AAAI Fall Symposium: From Reactive to Anticipatory Cognitive Embodied Systems.
Mahdavi, S. H. & Bentley, P. J. (2003). An evolutionary approach to damage recovery of robot motion with muscles. In Seventh European Conference on Artificial Life (ECAL03), 248-255, Spinger, Berlin.
Mirletz, B., Quinn, R. D. & SunSpiral, V. (2015). CPGs for Adaptive Control of Spine-like Tensegrity Structures, To Appear in Proceedings of 2015 International Conference on Robotics and Automation (ICRA2015) Workshop on Central Pattern Generators for Locomotion Control: Pros, Cons & Alternatives, Seattle, Washington, USA.
Srinivas, S. (1977). Error Recovery in Robot Systems . Ph.D.thesis, California Institute of Technology.