Session V64: Robophysics: Robotics Meets Physics II

Microrobots as robophysical models for the study of biomechanics and control of small animals

This talk will highlight our recent work on the development of physical models for the exploration of structure-function relationships in small animals (primarily arthropods), and for the development of robots that exhibit similar capabilities. Examples include centimeter-scale legged robots that help uncover gait strategies for high speed and robust locomotion on planar surfaces and for vertical and inverted climbing; ultra-fast power amplification mechanisms that produce rapid strikes and jumps; and insect-like flapping-wing robots used as a testbed for studies of fluid mechanics and under-actuated flight control. These models are enabled by the use of a multi-scale, multi-material fabrication paradigm, high bandwidth micro actuators, and detailed analytical, numerical, and experimental investigations. Robot complexity (e.g., measured by actuated degrees of freedom) typically decreases with reduced size. Our methods, however, buck this trend and allow us to create fully-actuated physical models that mimic key features of the biomechanics of the organisms in question. Furthermore, these robots serve as platforms for experimentation with novel sensors, computation architectures, and power solutions that all must reconcile strict size, weight, and power limits of these bioinspired devices with the desire to achieve similar capabilities as the organisms they are inspired by.   

Physics of animal and robot locomotor transitions in complex terrain

Robots are still poor at traversing complex terrain like earthquake rubble and construction sites for applications like search and rescue and structural examination. By contrast, animals move through nature by transitioning between different forms of movement. This is largely due to a lack of understanding of the physics of locomotor transitions in complex terrain. Here, we propose locomotion energy landscape as a framework for understanding how macroscopic, self-propelled, legged locomotors physically interact with terrain to probabilistically transition between locomotor modes. By integrating animal and robophysical experiments and physics modeling, we discovered that: (1) Different locomotor modes have different “terradynamic favorability”, measured by the potential energy barrier that must be overcome to traverse using each mode; and (2) Kinetic energy fluctuation from body vibration, induced by seemingly wasteful, oscillatory leg movement, helps animals and robots overcome mode-separating potential energy barriers and transition from less favorable to more favorable modes. Our study is a step in establishing terradynamics of locomotion in complex 3-D terrain. Such physics-based terrain traversal complements geometry-based obstacle avoidance and expands the reaches of robots.   

Giving Fish Robots a Pulse: Implementing Bio-inspired Control Algorithms in Fish Robots

Most fish use undulatory locomotion to control forward swimming and direction change relying heavily on their body and caudal fin to produce the motions. This style of locomotion allows for robust locomotion across aquatic environments. As such they have been models for designing autonomous under water vehicles (AUVs). To date, fish robots have been designed to maneuver using two basic modes of turning. The first is a waveform offset in which the frequency and amplitude of the oscillation remain unchanged, but the entire wave is biased to the right or the left, causing the fish to favor bending in one direction. The second is akin to the C-start maneuver in fish, in which a maximum amplitude deflection is simultaneously applied to all joints of the body to one side, interrupting the typical locomotor body oscillations. We developed a new turning model based on observations from high speed videos of live Giant Danio (Devario aquepinnatus). These videos show that maneuvers consist of propagating pulses of curvature. These pulses are independent events and can be modeled as a transient wave with a speed, amplitude, and width. Using a 3D printed robot, we will be evaluating the performance of the pulse model alongside the offset wave and C-start methods.   

Classroom Robophysics: Methods for teaching bioinspired design

In our Bioinspired Design course, students work in interdisciplinary teams to design engineering solutions based on biological analogies. Teams study core physics principles that enable biological functions and build prototypes to mimic these principles for applications in wearable technologies and robotics. Lectures cover current research in the area of bioinspiration, and the different design concepts and tools that can be used for bioinspired design. In this paper, we detail the approach the students undertake to distil the core physics principles and how to use them to create mechanical prototypes. The ultimate goal of the prototypes is to either solve an engineering challenge using lessons from nature, or to use engineering tools to study a biological phenomenon. In the course we present pedagogical tools to help students compare multiple organisms with similar functions but different underlying physics. Students are then asked to brainstorm different embodiments based on learned physics principles. They then apply design evaluation tools to choose the most suitable embodiment to solve the problem they identified at the start of the course. Students use their prototypes to verify that they captured the enabling physics rather than simply copying the superficial function.   

Fabricating Autonomous Machines for the Cellular Scale

This talk presents a new approach for fabricating cell-sized robots that can explore their environment, be manufactured en masse, and carry the full power of silicon-based information technology. We fabricate 10 to 100 um walking robots that are powered and controlled wirelessly using embedded silicon photovoltaics. Our robots walk using a new class of voltage controllable, electrochemical actuators made from nanometer thick membranes of platinum. These actuators impose low power requirements yet can carry loads ten thousand times their own weight. Moreover, actuation only requires 200 mV signals, facilitating straightforward integration with silicon microelectronics. Combine, these results present a broad platform that can unite mechanical systems, information processing and control into autonomous robots that operate at the cellular scale.   

Locomotion of Soft Robots with Flexible Flagella in Granular Medium

The solid friction analog to the resistive force theory (RFT) in viscous fluids can be applied in the context of granular media to describe the motion of sand lizards and snakes. Inspired by buckling in bacterial flagella in low Reynolds flow, we exploit the deformation of elastic structures for propulsion of limbed robots in granular medium. The centimeter-sized robots that are comprised of a rigid head and multiple elastomeric rods - our analog for flagella. Clamped at one end, the rods are rotated by a motor embedded in the head. This rotation generates a propulsive force that moves the entire robot forward. We combine Discrete Elastic Rods method with RFT to simulate the locomotion of this system. Side by side with simulations, we fabricate soft robots with wireless control and analyze their locomotion in a tank filled with clear beads. The combination of experimental and simulation data allows us to quantify the large deformation in elastic flagella resulting from the drag of granular medium. Finally, flagella of different geometrical shapes are explored to optimize the efficiency of the robot. This model robotic system can potentially help us understand the locomotion of living systems, e.g. bacteria.   

How to force an army of self-propelled mindless robots to act collectively?

We study assemblies of rodlike robots made motile through self-vibration. When confined in circular arenas, dilute assemblies of these rods act as a 2D gas of molecules. But above a critical surface fraction, some fraction of the bots line up in one or more tight clusters along the corral boundary while, in the bulk, gaslike behavior is retained. We find that the unified pushing of the cluster bots can drive collective motion: by selecting corrals that are deformable but free to move, we take advantage of surface cluster formation to force the robot army to work together.   

Gait coordination in bioinspired quadriflagellate robots

Multi-legged animals exhibit several distinctive patterns of limb movements, or gaits. In most cases, rhythmic patterns of limb actuation are generated by neural circuits called central pattern generators (CPGs), which operate in the absence of external timing cues or higher-level input. However, the capacity to coordinate locomotion gaits is by no means a feature exclusive to vertebrates. Indeed, species of micron-sized, pond-dwelling algae were recently discovered to be capable of orchestrating the beating of their four whip-like flagella to produce swimming gaits reminiscent of the motor patterns of quadrupeds (Wan & Goldstein 2016). Here it is thought that coordination is driven by contractile elements within the algal flagellar apparatus, which fulfil the role of the vertebrate CPG. In order to understand this unique intracellular control of motility, we developed robots which modeled quadriflagellate swimming at low-Reynolds number, and systematically evaluated the hydrodynamic performance of distinct gaits, including the trot, pronk, and gallop. Our results suggest a novel role of the algal cytoskeleton in providing mechanical stability during active flagellar beating.   

Body compliance helps snake robots traverse large steps

Snake robots still struggle to traverse complex 3-D terrain such as earthquake rubble and construction sites. By contrast, snakes traverse similar terrain like mountains and forests at ease. In both cases, how well the body engages the terrain is critical to successful traversal. Here, we use robophysical experiments to test the hypothesis that body compliance helps better engage and traverse complex 3-D terrain. We developed a snake robot with one-way wheels capable of traversing a large step using a partitioned gait that we recently discovered in snakes. An adjustable suspension was added to the wheels to vary compliance between the body and terrain. When traversing steps as high as 35 ± 3 % body length (BL), higher compliance allowed the suspension to compress more (from 1.0 ± 0.3 mm to 2.1 ± 0.6 mm) to better maintain contact with ground below and above the step (from 81 ± 8 % to 87 ± 5 % of the body), which increased traversal probability (from 63 ± 19 % to 90 ± 0 %) at the cost of 12 ± 6 % more power consumption (P < 0.05, ANCOVA). For larger step height (40% BL), however, the larger compression increased the chance of the rigid body edges being caught by the step corner. This is a limitation of the discrete robot body and highlights the need for a continuum compliant body.   

A robophysical model for studying obstacle navigation in sidewinders

Sidewinder rattlesnakes (Crotalus cerastes) use multiplane undulations to control their novel form of locomotion. Previous research showed that modulating amplitude or phasing of a two-wave control template—one controlling vertical lifting, the other lateral undulations—recapitulated turning gaits observed in the animals (Astley et al, PNAS 2015). With data from sidewinders interacting with pegs embedded in sand and a robophysical model, we investigate a peg-snake interaction where the animal forms a static contact with the substrate at an anatomical point anterior to the peg. This interaction mode facilitates “squeezing” the posterior body sections past the peg. Using a copper foil wrapped peg and 7 capacitive sensors along the 14-joint robot, we tested passive and active obstacle-clearing control templates. Early data shows the robot actively clears a peg using an increase in lateral wave amplitude or a reversal turn. When setting the multiplane undulations to produce two wavelengths along the robot’s body, the robot passively clears the peg if the peg interacts with a region of the robot’s body closest the head. These results will generate hypotheses for control strategies based on muscle activation patterns involved in sidewinding (Jayne, J. exp. Biol. 1988).   

A Robophysical Investigation of Series-Elastic Flapping Wings

Flying insects may achieve energy efficient flight by storing and releasing elastic energy in their thorax and muscle. Similarly, flapping wing micro-aerial vehicles (FWMAVs) may benefit from inclusion of elastic components in their actuation system. Despite significant investigation into the aerodynamics of flapping wings, the actuation of these movements through elastic structures in insects and robots is relatively unexplored. We have developed a dynamically-scaled robophysical experiment to study the dynamics of series-elastic flapping wings, with specific emphasis on discovering the role of linear and nonlinear elastic components in energy efficiency, perturbation resistance, and control. We vary system (elasticity) and actuation (amplitude and frequency) parameters and find that energy storage and recovery by an elastic element is dependent upon the stiffness of the element and upon the driving amplitude, frequency, and stroke profile. System response experiments suggest that the inclusion of series-elastic elements may have a negative overall effect on control capabilities. The results of the project will inform the design of future FWMAVs, providing insight into elastic element selection, power requirements, and control design.   

Viscous friction-like relationship arises from a simple Columb friction locomotion model

It is well known in thermodynamics that viscosity-like dissipation relationships appear in particle models even after only a small number of collisions was simulated. We present a minimalistic model of a robot moving inchworm-like in a Columb friction regime, which exhibits a similar effect. This model is sufficiently simple to allow the relationship between actuation force and asymptotic average speed to be solved in closed form. For the range of parameters relevant for decimeter to meter length robots, the relationship of force and speed that arises is nearly linear - suggesting that the average behavior of this locomoting system appears similar to pushing through a viscous medium, and that viscosity-like models might be more appropriate to robots than we previously assumed.   

Learning Multi-agent Workload Distributions in Confined Excavation

Our recent work with collective fire ant tunnel excavation revealed that approximately 30% of the workers performed 70% of the digging [Aguilar et al, Science 2018]. Complementary robot swarm experiments and numerical/theoretical models demonstrated the importance of this unequal workload distribution strategy in mitigating clogs and jams in congested environments. This contrasts with a strategy for spacious tunnels, in which maximal robot activity enables high performance. We have now augmented the robots with contact sensing capabilities so that each robot can distinguish different kinds of contact; preliminary results suggest that the duration and timing of robot-robot contacts correlates with tunnel density. We hypothesize that with the ability to deduce local tunnel density via contact sensing, the robots can excavate more effectively by adjusting their behavior transition probabilities based on their sensed environmental information. By varying the learning rules that rely on the local dynamics of the environment, we posit that the agents can learn transition probabilities to achieve an optimal digging performance. We explore the extent to which a learning-based approach can generate strategies that are robust to external and internal disturbances.