Monday, June 22
09:00 - 09:10 | Welcome remarks
09:15 - 10:15 | Keynote
Marco Santello, School of Biological and Health Systems Engineering, Arizona State University
"Control and adaptation of dexterous manipulation: Integration of predictive and reactive sensorimotor mechanisms."
Motor control and adaptation rely on complex interactions between reflexes and anticipatory control mechanisms. Through repeated exposure to mechanical interactions with the environment, the sensorimotor system learns to expect sensory consequences arising from motor actions, whereas reflexes would intervene when a discrepancy occurs between expected and actual sensory feedback. We have tested this theoretical framework in the context of grasping and dexterous manipulation using tasks that allow subjects to explore and choose relations between digit forces and positions. We have proposed that successful execution of manipulation of object grasped at self-chosen contacts likely requires prediction and sensing of digit placement for modulating digit force distribution. In contrast, grasping at the same constrained contacts can rely on sensorimotor memories of digit forces built through previous manipulations. We have tested this proposition by using behavioral experimental approaches to identify mechanisms underlying learning and execution of dexterous manipulation, and non-invasive brain stimulation to determine the role of specific brain areas within the cortical grasp network. I will review experimental evidence supporting the following notions: (1) interactions with objects grasped at unconstrained versus constrained contacts are mediated by different sensorimotor mechanisms, and (2) high-level representations acquired through sensorimotor adaptation allow the CNS to compensate for motor variability. I will conclude my talk with an overview of potential applications of this work to robotic grasping and prosthetics.
10:15 - 10:20 | Break / Refeshments
10:20 - 12:00 | Session 1A:
Tom Roberts, Ecology & Evolutionary Biology, Brown University
'Exploring the role of extracellular matrix in muscle mechanical performance'
Skeletal muscles are complex composites of cells containing contractile elements and extracellular connective tissue. The connective tissue serves to binds cells together, transmit forces, and resist muscle stretch at long lengths. Most studies of muscle contractile function use single cells with minimal extracellular connective tissue, and our understanding of muscle function is built on the mechanical behavior of these isolated elements. We hypothesize that the extracellular matrix plays an essential role in modulating muscle shape changes during contraction, and influences muscle force and speed via its influence on the trajectory of muscle fibers. Measurements of shape changes in muscles suggests that the spring-like behavior of extracellular matrix may have a significant impact on the flow of mechanical energy between contractile elements and the environment. Such observations suggest that viewing muscles as "scaled-up sarcomeres" may limit our understanding of the mechanics of muscle function during movement
Jonas Rubenson, Muscle Function & Locomotion Lab and Biomechanics Lab, Pennsylvania State University
"Distribution and modulation of joint mechanical work and power in bipedal locomotion"
Terrestrial legged animals depend largely on multi-articulated limbs for generating the mechanical work and power for locomotion. Understanding how terrestrial animals distribute work and power between their various joints, and how joint mechanics is modulated across locomotor tasks and environments can prove valuable for engineers designing articulated limbs for machines (e.g. robotics). Here we present findings from locomotor experiments in both mammalian and terrestrial avian bipeds. A general strategy among these species seems to be emerging: 1) a reliance on distal joints for providing the majority of positive mechanical work and instantaneous power for forward propulsion, 2) a reliance on the hip for limb protraction during swing phase, and 3) the reliance on the knee for energy dissipation. As speed and center of mass acceleration increases, distal joints contribute proportionately more to powering horizontal/vertical propulsion. Despite general trends, anatomical specializations result in differences in joint work and power distribution. For example, digitigrade animals exhibit much greater reliance on the metatarso-phalageal joint for powering locomotion whereas humans (plantigrade) exploit the foot’s arch elasticity. Future studies examining joint work across varied locomotor tasks including locomotion on different substrates and topographies will prove important for more fully understanding the mechanics and control of bipedal gait and its application to the adaptive motion of machines.
Monica Daley, Comparative Biomedical Sciences, Royal Veterinary College
"Investigating how animals integrate sensory information during adaptive locomotion"
Over the past several years, we have studied how avian bipeds (ground birds) move over uneven terrain in a number of different experimental conditions. An emerging theme is that, perhaps unsurprisingly, locomotor strategies are sensitive to the visibility of the upcoming terrain, and sensitive to recent terrain history. If a running bird can clearly see on obstacle or pothole well ahead of encountering it, it uses a safer and more economical strategy compared to negotiating a less visible or suddenly appearing terrain feature. Additionally, if terrain has recently been rough, birds tune their steady state gait to be more robust to perturbations. We are interested in understanding how locomotor control strategy is tuned based on sensory information, such as vision, proprioception, balance and inertia sensing. However, this is a challenging area of research, in part because it is not possible to directly measure most sensory variables of interest from animals during locomotion. While we can sometimes manipulate sensory information (for example, using virtual reality), we cannot be certain whether this manipulation is sensed as a realistic signal, or disregarded as a ‘noisy’ sensory channel. Animals sometimes respond to sensory perturbations in complex and surprising ways. I will discuss some of the approaches we are using, including virtual reality, surgically reduced proprioception, and vibrational stimulation of sensory afferents, and the current limitations of these approaches. I am interested in discussing advantages and disadvantages of different methods for investigating the role of sensory information in adaptive animal locomotion, with the goal of identifying fruitful collaborative questions for biologists and engineers. It may be the case, however, that the most appropriate methods for addressing the biological questions aren’t ideal for transferring principles to bio-inspired robotics. The sensory and neural control systems of animals differ markedly from the analogous systems in robots, and therefore may have different operating constraints. For example, animals suffer from neural delays that amount to a substantial fraction of a stride cycle, a limitation that machines do not have. Additionally, sensory systems are highly conserved in animal lineages, and therefore may not be optimal. One question open for discussion is whether there is a suitable level of abstraction for addressing the roles of sensory systems in animal locomotion that is usefully applicable in bio-inspired robotics. I will present some of my thoughts on this issue while aiming for a fluid and interactive discussion.
Dagmar Sternad, Departments of Biology, Electrical and Computer Engineering and Physics, Center for the Interdisciplinary Research in Complex Systems, Northeastern University
"Control of intermittent and continuous interactions with objects"
How do humans manage their actions and interactions with the physical world? How do we learn new skills or re-learn basic behaviors after injury, such as reaching to drink from a glass without spilling? Our research approach analyzes how task dynamics constrain and enable actions and their improvement with practice. Characteristic of our research is to start with a mechanical model of the task and render it in a virtual environment with a fully known workspace. Based on mathematical analyses of the modeled task, we study how humans develop stable solutions to meet complex task demands. Key concepts in our analysis are variability, stability, and predictability. Using three model tasks, throwing a ball, rhythmic bouncing of a ball, and transporting a “cup of coffee”, we show that humans develop skill by: 1) exploiting solutions with dynamic stability, 2) finding error-tolerant strategies and channeling noise into task-irrelevant dimensions, 3) optimizing predictability of object dynamics. These findings are the basis for developing propositions about the controller.
Hartmut Geyer, Robotics Institute, Carnegie Mellon University
"Decentralized control approaches to behavior adaptation in legged robots and powered prostheses based on neuromuscular models of human locomotion"
Neural networks along the spinal cord contribute substantially to generating locomotion behaviors in humans and other legged animals. Using computational models of the human neuromuscular system, we find that a spinal network based on muscle reflexes is sufficient to compose steady and transitional locomotion behaviors from walking and running, to slope and stair negotiation, to turning and deliberate obstacle avoidance. The results not only may help to better understand human motor control of locomotion, but also suggest decentralized control algorithms for adaptive behavior in legged machines. Exploring this second point, we highlight two application domains, humanoids and artificial limbs. We present our current results toward virtual neuromuscular controllers of bipedal robots that adapt gait to environment changes, and toward reactive controllers of powered knee prostheses for balance recovery and fall prevention in locomotion of transfemoral amputees.
12:00 - 13:15 | Lunch
13:20 - 14:50 | Session 1B:
Mark Cutkosky, Mechanical Engineering, Stanford University
"Adhesive scaling: from micro-tugs to macro-climbers"
Recent advances in understanding how to control and deploy gecko-inspired adhesives are enabling new applications at both small and large dimensional scales. At small length scales, "micro-tugs" exploit adhesion to interact with the world exerting forces many times greater than would be possible using frictional contact. In this application the behavior of directional adhesives has important implications for micro-tug actuation and power transmission. At very large scales, there are many options for actuating attachment and movement. The challenge, instead, is to maintain high adhesive pressures over large areas of contact. Once again, the behavior of directional adhesives has important implications for design of load application systems.
David Remy, Robotics and Motion Laboratory, University of Michigan
“Gaits and Natural Dynamics in Animals and Robots”
In our work, we seek to systematically exploit mechanical dynamics to make future robots faster, more efficient, and more agile. Drawing inspiration from biology and biomechanics, we design and control robots whose motion emerges in great part passively from the interaction of inertia, gravity, and elastic oscillations. Energy is stored and returned periodically in springs and other dynamic elements, and continuous motion is merely initiated and shaped through active actuator inputs. Within this context, this talk will focus specifically on the role of different gaits in multi-legged robots. In our interpretation, gaits are simply different `oscillation modes’ of the natural dynamics of the mechanical structure. We discuss very simple models of locomotion that can accurately predict a wide range of bipedal and quadrupedal gaits found in nature, and we show that the choice of gait can make a big impact on the performance of legged robotic systems. In the process, we shed more light onto the question what gaits actually are and discuss how they can be exploited by animals or robots.
Neville Hogan, Mechanical Engineering, MIT
"Managing physical interaction via dynamic primitives"
Despite vastly slower actuation and information transmission, animals and humans achieve agility far surpassing modern robots. Highly dynamic performance despite control and communication delays implies heavy reliance on feedforward control. That, in turn, requires ‘internal models’ of the body’s neuro-mechanics, and structures such as the cerebellum have been implicated in this function. But humans, in particular, excel at using tools and manipulating complex objects—consider fly-casting or cracking a whip. Using engineering models to predict the response of these latter objects challenges even the fastest modern computers. What form might biological ‘internal models’ of these dynamic objects take? I will explore the likelihood that biological feedforward control is based on dynamic primitives. Robot motion control using discrete and rhythmic motion primitives is well established. How to manage physical interaction is less clear. I propose mechanical impedances as a distinct class of dynamic primitives used for this purpose. I will discuss some of their required properties and how they may be integrated with motion primitives to achieve robust high-performance control.
Shinsuke Nakashima, University of Tokyo
"Balancing by Musculoskeletal Humanoid Kenshiro with Acquiring Muscle-COG Jacobian and Muscle Classification"
Animals have complex body structure and change in model throughout the growing process, but are still capable of controlling their bodies. From the thought that the Jacobian between whole-body and each muscle should be acquired by a robot itself, we propose a biomimetic method. The strategy is composed of three parts: self-acquisition of Jacobian with driving each muscle actuators, classification of muscles using the Jacobian, and reference distribution of each muscle's displacement based on the classification. The method was validated by balancing motion of musculoskeletal humanoid robot Kenshiro which has approximately 100 actuators.
Benjamin Robertson, Temple University
"Resonance in biological muscle-tendon as a mechanism for elastic limb behavior"
The fields of terrestrial biomechanics and bio-inspired robotics have identified spring-like limb mechanics as critical to stable and efficient gait. In biological systems, distal muscle groups cycling large amounts of energy in series tendons are a primary source of compliance. To investigate the neuromechanical origins of this behavior, we coupled a biological muscle tendon to a feedback controlled servomotor simulating the inertial/gravitational environment of terrestrial gait. This study concluded that by matching stimulation frequency to the oscillation frequency of the passive biomechanical system, muscle-tendon interactions resulting in spring-like behavior occur naturally and do not require closed- loop neural control.
14:50 - 16:30 | Refeshments/ Poster session 1 (Odd Numbered Submissions)
15:05 - 18:00 | Session 1C :
Andre Seyfarth, Institute of Sport Science, Technical University of Darmstadt
"Transferring biomechanical models to robotic and assistive devices: insights and benefits"
In this work we are going to outline a research approach for investigating concepts on legged locomotion with the help of bio-inspired hardware systems such as legged robots and prosthesis. We will explain why hardware models of biomechanical gait concepts are not just a nice feature but do also provide a scientific tool to demonstrate and prove the value of the conceptual insights. Hardware models share properties of conceptual models (being simplified and human-made) and of the biological system (being realistic). Thus they provide a valuable column of the bridge connecting real biological locomotion from conceptual computer simulation models and theories on legged locomotion. With assistive devices this approach can be taken one step further, namely by applying the constructed concepts to the human body and by investigating the interactions between human movement and constructed concept of this very same human motion.
Jonathan Hurst, Mechanical Engineering, Oregan University
"Walking and Running with ATRIAS, a Human-Scale Bipedal Robot"
ATRIAS, our human-scale bipedal robot, is the first machine to reproduce human-like and animal-like ground reaction forces and center-of-mass motion for a bipedal walking gait. This result is due to two key components: First, the machine is designed to exhibit carefully designed passive dynamics based on spring-mass models of animal locomotion. Leg springs, in series with powerful actuators, are mechanical and not implemented through control. They are sized specifically to generate oscillations at the frequency of the gait, and to store gait energy; this is a key difference from many series elastic actuators. Second, the control methods are designed explicitly to rely on the passive dynamics of the machine, and to integrate and work with the spring-mass behavior to generate desired behaviors for agile, efficient, robust walking and running. In this talk, I will discuss the challenge of handling ground impacts and state-dependent controller transitions, as well as our methods of energy regulation, speed regulation, and balance control. I will report on experimental results with ATRIAS, and show videos of ATRIAS walking over flat and rough ground.
Jonathan Clark, Mechanical Engineering, Florida State University
"Robotic Limb Design for Dynamic, Multi-purpose Behaviors"
Finely tuned robotic limb systems that explicitly exploit their body’s natural dynamics have begun to rival specific performance criteria, such as speed over smooth terrain, of the most accomplished biological systems. The earliest successful robot implementations however, used only very specialized designs with a very limited number of active degrees of freedom. While more flexible, higher degree-of-freedom designs have been around for some time they have usually been restricted to comparatively slow speeds or manipulation of light-weight objects. The design of fast, dynamic multi-purpose robots has been stymied by the limitation of available mechanical actuators and the complexity of the design and control of these systems. This talk will describe recent efforts to understand how to effectively exploit available actuator power and to judiciously add complexity to the design of multi-use limbs to enable dynamic motions in multiple modalities and for dynamical mobile manipulation.
Shai Razven, Electrical Engineering and Computer Science, University of Michigan
"Gait Dynamics as Hybrid Oscillators"
Animal gaits can be viewed as rhythmic body motions which propel the animal with respect to the surrounding substrate. In terrestrial legged locomotion, these can often be represented as periodic solutions of a hybrid dynamical system – a collection of flows spliced together at surfaces representing discontinuities that arise when contact with the ground is established or broken. While hybrid system models have been in use by biomechanists and robot designers for over 30 years, no special attention has been given to hybrid oscillators in particular. I present recent theoretical results suggesting that hybrid oscillators offer unique advantages for control in comparison with similar smooth (non-hybrid) oscillators. I also present the application of Data Driven Floquet Analysis to human running, allowing the construction of predictive models with unprecedented accuracy. The hybrid oscillator perspective on gait dynamics provides a coherent experimental and theoretical framework, allowing insights from animal locomotion to be represented mathematically and carried across into robotic applications.