SNF Project Locomotion: Final report 2009-2010

2000

The sensory control of mammalian locomotion has been studied for around 150 years. Many systems are involved: skeletomuscular actuators, spinal reflex and pattern generating networks, propriospinal and brainstem networks, the vestibular apparatus, cerebellum, deep brain nuclei and the cerebral cortex. All of these systems are directly or indirectly affected by mechanical or sensory input related to the locomotor movements they help control. In this chapter we will argue that since locomotor movements vary tremendously according to task, terrain and context, sensory input is crucial in controlling them. Experimental results as well as control systems simulations will be used to show that sensory input is crucial for determining the timing of phase transitions and thus the cadence (cyclic frequency) of locomotion, as well as the relative scaling of muscle activation and thus the amplitude of locomotor movements. We will argue that stretch reflexes play a relatively minor role in this scheme, basically augmenting the load compensation that occurs by virtue of the intrinsic compliance characteristics of skeletal muscle. With the help of biomechanical models, we will show that in quadrupeds, a carefully constructed pattern of muscle activations can produce sustained locomotion over flat ground in the absence of sensory input. However, small variations in terrain, initial conditions, or biomechanical parameters can disrupt locomotor stability. The situation is more critical in bipedal locomotion, where sensory input is vital for stable step cycles. The main function of the huge flow of multimodal sensory information from limb mechanoreceptors to the CNS is therefore continuously to adapt the locomotor pattern to variations in terrain and posture and to mediate higher level prediction of requirements in upcoming step cycles. Sensory systems must therefore be considered as integral parts of the semi-autonomous locomotor pattern generator. Taking the broad view, at least five levels of locomotor control can therefore be identified: load compensation due to skelotomuscular properties, load compensation due to stretch reflexes, bodily motion resulting from cyclical pattern generation and adaptive and predictive control in relation to terrain and to behavioural goals and contexts.

HART 2004, Fourth Intl. Symp. on Human and Artificial Intelligence Systems: From Control to Autonomy, 2004

Abstract. Why is walking easy for us and difficult for robots? The conceptual framework of traditional Artificial Intelligence, and constraints from the background of industrial robot designers, means that commercial humanoid design has often ignored the role of natural dynamics of a mechanical system, as illustrated in a Passive Dynamic Walker. We show, using a Dynamical Systems approach and Evolutionary Robotics, how power and control can be added to such systems, demonstrating bipeds in simulation with many degrees of ...

Journal of Intelligent & Robotic Systems, 2014

In this paper we review more than 140 publications and try to not only give a snap shot of the current state of the art research in the area, but also to critically analyse and compare different methodologies used in this research field. Among the investigated intelligent approaches for solving locomotion problems are oscillator based Central Pattern Generators, Neural Networks, Hidden Markov models, Rule Based and Fuzzy Logic systems, as well as Analytical concepts. We try to compare those methods based on the quality of the produced solutions in terms of time, stability, correctness and the expense and cost for achieving them. At the end of each section we list the advantages and disadvantages of the reviewed methods and scrutinize them considering the complexity of the approaches, their applicability to the investigated locomotion tasks and the constraints of the produced solutions. The reviewed publications examine a range of legged and non-legged systems, operating in simple and complex environments, with several different locomotion tasks.

Journal of …, 2002

From what we know at present with respect to the neural control of walking, it can be concluded that an optimal biologically inspired robot could have the following features. The limbs should include several joints in which position changes can be obtained by actuators across the joints. The control of mono-and biarticular actuators should occur at least at three levels: one at direct control of the actuators (equivalent to motoneuron level), the second at indirect control acting at a level which controls whole limb movement (flexion or extension) and the third at a still higher level controlling the interlimb coordination. The limb level circuits should be able to produce alternating flexion and extension movements in the limb by means of coupled oscillator flexor and extensor parts which are mutually inhibitory. The interlimb control level should be able to command the various limb control centers. All three control levels should have some basic feedback circuits but the most essential one is needed at the limb control level and concerns the decision to either flex or extend a given limb. The decision to activate the extensor part of the limb oscillator has to be based on feedback signalling the onset of loading of the limb involved. This should be signalled by means of load sensors in the limb. The decision to activate the flexor part of the limb oscillator has to depend on various types of feedback. The most important requirement is that flexion should only occur when the limb concerned is no longer loaded above a given threshold. The rule for the initiation of limb flexion can be made more robust by adding the requirement that position at the base of the limb (''hip'') should be within a normal end of stance phase range. Hence, human locomotion is thought to use a number of principles which simplify control, just as in other species such as the cat. It is suggested that cat and human locomotion are good models to learn from when designing efficient walking robots.

… of the Royal …, 2007

Research on the biomechanics of animal and human locomotion provides insight into basic principles of locomotion and respective implications for construction and control. Nearly elastic operation of the leg is necessary to reproduce the basic dynamics in walking and running. Elastic leg operation can be modelled with a spring-mass model. This model can be used as a template with respect to both gaits in the construction and control of legged machines. With respect to the segmented leg, the humanoid arrangement saves energy and ensures structural stability. With the quasi-elastic operation the leg inherits the property of self-stability, i.e. the ability to stabilize a system in the presence of disturbances without sensing the disturbance or its direct effects. Self-stability can be conserved in the presence of musculature with its crucial damping property. To ensure secure foothold visco-elastic suspended muscles serve as shock absorbers. Experiments with technically implemented leg models, which explore some of these principles, are promising.

While internal models are a prerequisite for higher-level function, they have to be grounded in lowerlevel function serving sensorimotor control. In this paper we introduce an internal body model for the control of a hexapod walker. The internal model deals with a highly complex robotic structure of 22 degrees of freedom and coordinates the single joint movements to achieve an overall stable and adaptive walking behavior. It is implemented as a hierarchical recurrent neural network consisting of different levels of abstraction which are tightly intertwined. We demonstrate the feasibility of the concept by applying the model to a simulated robot and show how the different levels of the body model interact and how this allows to scale the model even further. While the internal model is used in this context explicitly for motor control, it is also a predictive model and can be applied for sensor fusion. We discuss how in this way such an internal model offers the flexibility to be utilized in motor control and to be used for planning ahead by a cognitive expansion of the movement controller.

The sensory control of mammalian locomotion has been studied for around 150 years. Many systems are involved: skeletomuscular actuators, spinal reflex and pattern generating networks, propriospinal and brainstem networks, the vestibular apparatus, cerebellum, deep brain nuclei and the cerebral cortex. All of these systems are directly or indirectly affected by mechanical or sensory input related to the locomotor movements they help control. In this chapter we will argue that since locomotor movements vary tremendously according to task, terrain and context, sensory input is crucial in controlling them. Experimental results as well as control systems simulations will be used to show that sensory input is crucial for determining the timing of phase transitions and thus the cadence (cyclic frequency) of locomotion, as well as the relative scaling of muscle activation and thus the amplitude of locomotor movements. We will argue that stretch reflexes play a relatively minor role in this scheme, basically augmenting the load compensation that occurs by virtue of the intrinsic compliance characteristics of skeletal muscle. With the help of biomechanical models, we will show that in quadrupeds, a carefully constructed pattern of muscle activations can produce sustained locomotion over flat ground in the absence of sensory input. However, small variations in terrain, initial conditions, or biomechanical parameters can disrupt locomotor stability. The situation is more critical in bipedal locomotion, where sensory input is vital for stable step cycles. The main function of the huge flow of multimodal sensory information from limb mechanoreceptors to the CNS is therefore continuously to adapt the locomotor pattern to variations in terrain and posture and to mediate higher level prediction of requirements in upcoming step cycles. Sensory systems must therefore be considered as integral parts of the semi-autonomous locomotor pattern generator. Taking the broad view, at least five levels of locomotor control can therefore be identified: load compensation due to skelotomuscular properties, load compensation due to stretch reflexes, bodily motion resulting from cyclical pattern generation and adaptive and predictive control in relation to terrain and to behavioural goals and contexts.

Frontiers in Psychology, 2019

Features of gait are determined at multiple levels, from the selection of the gait itself (e.g., walk or run) through the specific parameters utilized (stride length, frequency, etc.) to the pattern of muscular excitation. The ultimate choices are determined neurally, but what is involved with deciding on the appropriate strategy? Human locomotion appears stereotyped not so much because the pattern is predetermined, but because these movement patterns are good solutions for providing movement utilizing the machinery available to the individual (the legs and their requisite components). Under different circumstances the appropriate solution may differ broadly (different gait) or subtly (different parameters). Interpretation of the neural decision making process would benefit from understanding the influences that are utilized in the selection of the appropriate solution in any set of circumstances, including normal conditions. In this review we survey an array of studies that point to energetic cost as a key input to the gait coordination system, and not just an outcome of the gait pattern implemented. We then use that information to rigorously define the construct proposed by Sparrow and Newell (1998) where the effects of environment, organism, and task act as constraints determining the solution set available, and the coordination pattern is then implemented under pressure for energetic economy. The fit between the environment and the organism define affordances that can be actualized. We rely on a novel conceptualization of task that recognizes that the task goal needs to be separated from the mechanisms that achieve it so that the selection of a particular implementation strategy can be exposed and understood. This reformulation of the Sparrow and Newell construct is then linked to the proposed pressure for economy by considering it as an optimization problem, where the most readily selected gait strategy will be the one that achieves the task goal at (or near) the energetic minimum.

2010 10th IEEE-RAS International Conference on Humanoid Robots, 2010

4