+ Thomas G. Sugar, Andrew R. Bates, Sangram Redkar, and Jason Kerestes, Using a Phase Oscillator to Control Oscillatory Behavior. [pdf][poster]
Brian T. Mirletz, Roger D. Quinn, and Vytas SunSpiral, CPGs for Adaptive Control of Spine-like Tensegrity Structures. [pdf][poster]
Thierry Hoinville, Malte Schilling, and Holk Cruse, Control of rhythmic behavior: Central and Peripheral Influences to Pattern Generation. [pdf][poster]
Dai Owaki, Shunya Horikiri, Jun Nishii, and Akio Ishiguro, Load-dependent Interlimb Coordination for Bipedal Walking. [pdf][poster]
Gaudi Morantes, Simon Guerrero, Jose Cappelletto, Gerardo Fernandez, and Rafael Torrealba, Comparison of CPG topologies for bipedal gait. [pdf][poster]
+ Nicholas S. Szczecinski, Alexander J. Hunt, and Roger D. Quinn, Phase Characterization of Hyperpolarizing and Depolarizing Currents on a Four Neuron CPG Model. [pdf][poster]
+ Vojtech Vonasek, Martin Saska, Utilization of CPGs in sampling-based motion planning. [pdf][poster]


Prof. Auke Ijspeert, Using robots and numerical models to decode spinal circuits controlling locomotion in vertebrate animals, from lamprey to human

  • The ability to efficiently move in complex environments is a fundamental property both for animals and for robots, and the problem of locomotion control is an area in which neuroscience and robotics can fruitfully interact. Animal locomotion control is in a large part based on spinal cord circuits that combine reflex loops and central pattern generators (CPGs), i.e. neural networks capable of producing complex rhythmic or discrete patterns while being activated and modulated by relatively simple control signals. These networks located in the spinal cord for vertebrate animals are modulated by descending control signals and interact with the musculoskeletal system for generating rich motor behaviors. In this talk, I will present how we use numerical models and robots to explore the interplay of these four components (CPGs, reflexes, descending modulation, and musculoskeletal system). Going from lamprey to human locomotion, a series of models will be presented that tend to show that the respective roles of these components have changed during evolution with a dominant role of CPGs in lamprey locomotion, and a more important role for sensory feedback and descending modulation in human locomotion. Interesting properties for robot locomotion control will also be discussed.
Prof. Art Kuo, Feedforward and feedback control of robot locomotion: The conflict and its resolution
  • Biology has provided inspiration for a variety of robots that walk, fly, or swim. Bio-inspiration can, however, take many forms, ranging from a general morphology such as the number of legs, to also specifying the control architecture. Perhaps the most such control architecture is the central pattern generator (CPG), which in biology refers to a neural oscillator that can produce rhythmic motor commands even in the absence of sensory input. This is typically interpreted to mean that the CPG drives locomotion, but this is theoretically problematic for stability. The principles of feedback control dictate that the optimal control should be driven by the robot’s states (e.g., positions and velocities of all joints). The addition of extra dynamics, in the form of the neural oscillator, has little place in that framework. In fact, any additional dynamics would typically be treated as a disturbance, which at best has no effect, and at worst destabilizes the gait. Here we will explore the implications of the CPG on feedback control and re-examine the CPG’s role in locomotion. We will examine a framework where the CPG constructively contributes to locomotion, but not as the driver of locomotion. Instead, the CPG may be interpreted as a crucial component within an estimator-based, state feedback controller.


Prof. Ioannis Poulakakis, A nonlinear control design approach to dynamically-stable running robots

  • Robotic legged locomotion is characterized by continuous dynamics punctuated by discrete contact events. Furthermore, the short time intervals within which corrective action must be developed and the limited capabilities of the actuators available — particularly of those used in small-scale legged robots — render the design of controllers for such systems a challenging task. One approach to designing controllers for legged robots employs neurally inspired control architectures to excite locomotion patterns on legged robots; central pattern generators (CPGs) are a representative example. An alternative approach to CPG-based controllers uses nonlinear feedback synthesis tools to coordinate the actuated degrees of freedom of a legged robot so that a lower-dimensional dynamical system emerges from the closed-loop robot dynamics; this lower-dimensional system effectively encodes the targeted locomotion behavior and governs the motion of the robot. This talk will focus on the application of nonlinear state-feedback design tools for locomotion control using a family of canonical locomotion models (templates), and will attempt to make connections with more bio-inspired control architectures such as CPGs.


Prof. Jean-Marie Cabelguen, Locomotion in limbed vertebrates: from CPGs to behavior

  • The patterned muscle contractions underlying locomotion in vertebrates are produced by neural networks located in the spinal cord (central pattern generators: CPGs). During the last decades, the use of different vertebrate models, novel neurobiological techniques and modelling has enabled considerable progress to be made in our understanding on the operating mode of the CPGs for locomotion. These studies further revealed that CPGs are continuously influenced by descending and sensory signals. These influences help the animal to rapidly adapt the limb movements to the external environment and change in gait. The present talk will attempt to give an overview of the current knowledge on the neural mechanisms underlying the flexibility in the operation of the locomotor CPGs in limbed vertebrates. The contribution of that flexibility to adapted locomotor behavior will be addressed.


Prof. Low Kin Huat, Optimization of biomimetic undulatory swimming by an experiment-based approach via CPGs

  • Inspired by a biological neural network, central pattern generators (CPGs) have been commonly considered as a group of coupled neurons that generate rhythmic signals. In the present study, an experiment-based approach is modeled to improve the performance of biomimetic undulatory locomotion of fish swimming through on-line optimization via CPGs. The approach is implemented through two phases. In the initial phase, coordinated swimming gaits are generated by CPGs. Next, optimal parameter sets for the CPG model is searched on-line by using Genetic Algorithm (GA). The effectiveness of the approach is demonstrated in the optimization of swimming speed and energy efficiency for a biomimetic fin (or body) propulsor. To evaluate how well the input energy is converted into the kinetic energy of the propulsor, an energy-efficiency index is presented and utilized as a feedback to regulate the on-line searching with a closed-loop swimming control. Experiments were conducted on propulsor prototypes with different fin segments and the optimal swimming patterns were found separately. The comparisons of results show that the optimal curvature of undulatory propulsor, which might have different shapes depending on the actual prototype design and control scheme. Some main issues arising from the design, optimization, and implementation of the CPG-based control are also discussed.


Prof. Hartmut Geyer, The functional role of CPGs in legged systems that require active balancing remains an open and compelling research problem 

  • CPGs simplify locomotion control if dynamic balance is not crucial. But the more an animal or legged robot depends on active balancing, the more its control will be governed by state feedback. As top-heavy bipeds, humans are quintessential examples of such a system, and it is unlikely that their locomotion control is generated by CPGs. To substantiate this notion, I will present in the first part of this talk a computational model of human neuromuscular control which only uses state feedback to generate diverse locomotion behaviors while producing plausible muscle activations and reflex responses. However, physiological evidence of CPGs in animal species from which we have evolved suggests that CPGs remain a component of the motor control network in humans. In the second part, I will use insights from the computational neuromuscular model to reflect on common ideas about how CPGs might integrate with state feedback control from feedback learning and state machines to internal clock drives and observers.


Prof. Dai Owaki, Prof. Akio Ishiguro, A minimal CPG model for interlimb coordination in quadruped locomotion

  • Quadrupeds exhibit flexible and versatile locomotion in real-world environments. Well-known experiments involving decerebrate cats have suggested that such locomotive patterns are partly controlled by an intra spinal neural network called the “central pattern generator” (CPG). However, the mechanism responsible for leg coordination remains elusive. Here, we present our minimal CPG model for quadruped interlimb coordination, where coordination between legs is self-organized only with the use of local load sensing, and without any pre-programmed patterns. By using an embodied quadruped robot, our model well explains a large range of observations of quadruped gait patterns. In this workshop, we would like to discuss its validity to biology, scalability to various body properties, etc. 


Dr. Victor Barasuol, Reactive Controller Framework: CPG-inspired trajectory generation, trajectory modulation and decision guidance

  • HyQ Robot is a 80 Kg hydraulically actuated quadruped robot that is a research platform able to walk and trot on flat and uneven terrains. To perform walking trot and flying trot, HyQ uses a reactive controller framework that has a CPG-inspired motion generation. In this talk I present a series of successful locomotion experiments performed by HyQ that involves directly or indirectly the idea of central pattern generators (CPG) for trajectory generation, easy parameter modulation and for communication between medium and high level tasks. I highlight the pros and cons that led to the current used approach, the trial results of different CPG output modulation by means of sensory feedback and the studies on CPG modulation to increase locomotion robustness. To conclude, I bring out the idea of using CPG states for decision guidance.


Dr. Mostafa Ajallooeian, Systematic design of custom oscillators to create high-level CPG models

  • There are different methodologies to create mathematical models which encode desired rhythmic profiles as their limit cycle, including network of Matsuoka oscillators, pool of adaptive frequecy oscillators, recurrent neural networks, etc. This talk is about the concept of Morphed Oscillators, which are able to encode arbitrary rhythmic profiles as limit cycles of nonlinear oscillators of desired orders while ensuring assymptotic stability. We show how such oscillators can be constructed in a very easy and elegant manner. Additionally, we show how sensory feedback can be integrated into such oscillators for the task of quadrupedal rough terrain locomotion.