Robots as Energy Systems: Advances in Robotics across Scales and Technologies

Robots are operating at unprecedented scales and in uniquely challenging environments, particularly near the human body. These robots are enabled by novel actuation, sensing, energy storage, and conversion technologies. Across different scales and between different technologies, the key metrics of performance are related on energy: how it is stored, delivered, or dissipated.


Introduction
Scientists and engineers across the world are exploring two key frontiers in Robotics: soft and small machines. Soft robotics promises of safe operation near the human body, potentially disrupting healthcare, manufacturing, agriculture and beyond. Miniaturized robots, operating alone or as swarms, will revolutionize search and rescue operations, surveillance, medical devices and space exploration. As devices get developed, the lines between small and soft become blurred, and comparing different technologies becomes increasingly difficult. For example, both soft and rigid actuators become capable of driving the same type of flapping wing milli-robot at the same scale. [1,2] In my research, I evaluate robots as energy systems, carefully tracking the input energy, efficiency of conversion and the energetic cost per task completed by the robot. The energy and power are normalized relative to the robot's mass (or volume), then specific energy or specific power (or energy density and power density) can be used to compare across technologies and across scales. The approach of evaluating robots as energy systems provides a framework to compare across scales, actuation technologies, energy storage mechanisms, or simply transducers in general. Alternatively, giving a full accounting of how many Joules of energy a robot starts with, and how many are used per task, may provide roboticists with an interdisciplinary perspective on how robots can be improved. Colleagues are already discussing how batteries could be scaled down [3] to power smaller robots and sensors, and how embodied energy can support autonomous systems. [4] The same discussion should happen for all robot components, in particular actuators, that are key drivers of performance in soft and small systems.

Discussion
Presenting the full picture of a robotic component's performance is typically not the norm: most authors look for ways to describe their technology as unique while downplaying the limitations of their approach. This makes direct comparisons challenging, and reduces technology adoption across different platforms and areas of robotics. However, understanding the current limitations is fundamental in targeting what to improve next. For example, in the field of dielectric elastomer actuators (DEAs), the early work by Pelrine et al. [5] showed enormous promise, particularly high strain (>100%) and energy density (3.4 MJ m À3 ). It also highlighted the limitations of their work, specifically the need for prestretching the elastomer, limited operating life, slow deformation, etc. The clear targets for improvement allowed for significant advances in the coming decades, with Duduta et al. [6] eliminating the need for prestretch, Lotz et al. [7] demonstrating improved cycle life (3 Â 10 9 cycles), and Maffli et al. [8] showing ultrafast deformation (175 μs settling time for a tunable DEA lens). As a community, we should embrace the approach of discussing energy performance of our robotic components to make cross-disciplinary research more impactful, to communicate our results more clearly to the general public, and to address societal challenges more effectively.
As individual researchers, we should strive to measure performance along clear, energy-focused metrics, and report it in a factual, straightforward way. For example, a new actuation technology that uses a phase changing material to do mechanical work should clearly present all of the inputs to the system that cause the thermally driven phase change, e.g., the input current and resistance, as well as estimates for the energy dissipated as heat to the environment. Work from Miriyev et al. [9] demonstrates a soft actuator in which a material undergoing a liquid to gas phase change while embedded in an elastomer can do meaningful mechanical work to the environment. The results are presented in rigorous detail, going through the details of the heating mechanism and conditions, to the mechanical output of the device, to arrive at an energy efficiency of the actuator of 0.2%. The exact value can be judged as high or low only in the context of an application, specifically if the robot has sufficient energy to complete its assigned tasks. Similarly, work by Duduta et al. [10] describes dielectric actuator performance showing both the highlights (20 J kg À1 specific energy, on par with natural muscles) and the lowlights (<0.1% energy efficiency). Several reviews delve into comparing different actuator types in terms of operating conditions, including Kim et al. [11] who bring a materials perspective, as well as El-Atab et al. [12] who focus more on robotic applications. However, describing the promise and capability of a specific technology at the same time as its limitations is the best way of ensuring the next generation of researchers can make fundamental advances.
Power autonomy is a key challenge for robotic swarms, machines operating in unstructured environments, and other emerging research areas. To achieve the goal of power autonomy, we need an interdisciplinary approach toward on-board energy sources, as well as harvesting and conversion mechanisms. Recent results from Jafferis et al. [13] demonstrate power autonomy for a flapping wing miniature robot (<500 mg) at an extremely challenging scale (<5 cm flapping wing span). From an energy perspective, incoming photonic energy from a radiative source is converted into electrical power by a solar cell array. The electric output of the solar cells is boosted to the necessary driving voltage for piezoelectric actuators (>190 V). The ceramic lead zirconia titanate (PZT) structures deform due to a piezoelectric effect, and vibrate at high frequency (>200 Hz). Through a flexure-based mechanism, the bending of PZT actuators is converted into flapping of lightweight wings, which cause the vehicle to take off. The work is a massive achievement for the field, and shows the level of complexity and interdependence of advanced robotic systems. Any demonstration of robotic power autonomy in any environment should aim to replicate this level of detail and consideration for each step in the energy conversion cascade. In the same research arena, James et al. [14] demonstrated liftoff of a 190 mg aerial vehicle wirelessly powered by a laser. Both of these groups demonstrate that energy accounting is extremely valuable, allowing the researchers to optimize the system where needed to ensure optimal operation of their robots.
The ability to store energy and deliver it as fast as needed for a specific task remains a significant consideration as devices are scaled down for micro-and even nanorobots. While millirobots can typically be assembled via laminate fabrication methods, [15] further reduction in scale requires either microfabrication with microelectromechanical system (MEMS) approaches [16] or wet chemistry techniques. [17] Recent work by Greenspun [18] shows robot made completely via MEMS techniques capable of leaping, as a mode of locomotion. The article carefully documents how electrostatic actuators convert an electrical input into mechanical work, which can be stored in submillimeter serpentine springs. In one example of self-powered locomotion, the rapid release of 1.05 μJ led to a 1 mm jump for a 6 mm tall robot. At an even smaller scale, Jin et al. [19] demonstrate a functional nanorobot capable of chemophotothermal therapy. When designing their robot, the authors considered multiple ways to power the device, including ultrasonic, chemical, magnetic, light, and even by a living system. All of these approaches have advantages at the nanoscale because the robot can harvest energy directly from the environment. For their specific application (e.g., navigation followed by drug release), the choice was magnetic propulsion coupled with near-infrared (NIR) light powered drug delivery.
The design and synthesis of the robot was based on the materials that enable energy conversion at the desired time in the nanorobot's operation. Even at the smallest scale, the fundamental principles that allow nanorobots to complete biomedical tasks are based on energy storage and conversion.
At some scales and in certain environments, carrying an on-board power supply is the only path to power autonomy. For robots that need to be compliant and adapt to their environment, the strain requirements on the entire robot can be satisfied by highly deformable rechargeable batteries. Kim and coworkers [20] demonstrate a stretchable (up to 90% of its initial length) and rechargeable (up to 100 cycles at 0.5C) battery that can undergo 36 000 mechanical deformation cycles with minimal impact on its electrochemical performance. The battery geometry uses the scales of a snake as bioinspiration for creating a deformable structure with rigid components connected by hinges. The researchers used kirigami and origami techniques to create hinges between rigid components that enclose the active battery materials and undergo no strain. The fabrication approach addresses both the electrochemical requirements, specifically the need to encapsulate the battery materials to prevent moisture from damaging the active materials, as well as the mechanical requirements, specifically the ability to allow for 90% strain across the entirety of the battery. The work concludes with a robotic demonstration, pairing the stretchable battery with a toy snake and a toy caterpillar to produce effectively an autonomous soft robot at the scale of interest. This recent work highlights the need for roboticists to collaborate with materials scientists and chemical engineers to find integrated solutions that address both mechanical and electrochemical challenges in new types of deformable and rechargeable batteries.
Alternatively, different robotic technologies may be developed around a specific energy storage solution that serves multiple functions. Drawing inspiration from natural systems, a single component can serve multiple functions, for example, a muscle can work as an actuator, a spring, or even a thermal generator in an emergency situation. Following the same concept, a project led by Prof. Kotov [21] has demonstrated batteries with configurable shape that can serve as structural elements in novel types of robots. The key materials are aramid nanofibers which can be made into composite structures to produce pliable zinc-air batteries. The highest performance battery shows a specific energy around 842 Wh kg À1 , a factor of 3-8Â higher than Li-ion batteries at a similar form factor, with the advantage of being a deformable structure. The work focuses heavily on the synthesis and characterization of the ceramic composites, and demonstrates the battery technology as the power supply for commercially available small toy robots. For both this zinc-air battery and the snake-inspired stretchable battery, a clear focus on a robotic problem and consideration for energy conversion throughout the robot's operation will enable exciting new areas of research.
Hybrid approaches can be extremely valuable, as evidenced by the work of Aubin et al. who developed "Electrolytic vascular systems for energy-dense robots". [22] In their demonstration, the researchers used a redox flow-type battery to drive fluidic actuators in a soft, energy autonomous, robotic fish. Effectively, the fluids served two purposes: energy storage liquids for the redox flow battery and working fluids for the deformable soft actuators that make the robot fish swim. The relative energy www.advancedsciencenews.com www.advintellsyst.com stored and used in each component is critical to extending the operating life of the robot. Specifically, the authors report a system energy density of 53 J g À1 , allowing the robot to operate continuously for over 36 h. Natural systems are a continuous source of inspiration, particularly muscles which serve a wide range of purposes, including operating as actuators, breaks, struts, springs, and even heating elements. [23] Other researchers may be inspired by this hybrid approach and deploy more multifunctional systems in which a robotic component serves a variety of functions. Mechanical energy can also be stored in deformable elements, and released rapidly for impulsive motions, such as projectile throwing or jumping. These power amplification structures can be deformed slowly and store the mechanical energy as strain energy in a deformable component. Through design, geometry or material selection, the deformable structure is made to have a rapid energy release mechanism, for example as a flexure with a hinge that can rapidly deform once the hinge is released. The slow energy uptake and fast release allows the robot designer to tailor the power delivered locally, by controlling the amount of elastic energy stored and how quickly it can be released.
Baek et al. [24] demonstrate a versatile energy release mechanism based on compliant origami that can be used for a gliding robot, a flapping wing mechanism, as well as a flea-inspired jumping robot. The fabrication method is supported by thorough mathematical modeling guiding the reader through the design process while focusing on the energy stored and speed or release, aiming to match the performance of ladybug beetles. The article encourages a discussion on pairing power amplification mechanisms with specific actuation technologies to meet requirements across different applications. A similar approach focused on energy amplification has enabled Steinhardt et al. [25] to build a physical model of the mantis shrimp and explore the dynamics of ultrafast systems. The same focus on energy and power should be applied when characterizing power amplification mechanisms, and understanding their role and efficiency in an integrated robotic system.
A denser form of energy storage is chemical fuels, which are used abundantly in transportation at large scales. As robots are miniaturized, the ability to use a chemical fuel efficiently becomes very limited. Internal combustion engines reach limits of friction due to scaling as they are reduced in size. Researchers need to develop alternative pathways to convert the energy dense fuels into useful mechanical work. Yang and co-workers [26] demonstrate an insect-scale (88 mg) autonomous crawling robot that is powered by methanol. The liquid fuel reacts with ambient oxygen on a catalyst, and the resulting heat causes deformation in a shape memory alloy (SMA) wire. The cycles of contraction and expansion of the SMA wire drive legged locomotion of the robot at around 0.0185 body lengths/sec. The total energy efficiency of the robot is estimated to be around 0.48% from chemical fuel to mechanical work, highlighting the inherent advantages of high energy density fuels, where even a low energy conversion efficiency (e.g., 0.2%) from a high-density source (e.g., 22.4 kJ g À1 for methanol) still corresponds to sufficient energy at the robot level (e.g., 44.8 J g À1 , on par with the robot fish developed by Aubin). Soft robots can also be paired with chemical fuels, as shown in work by Chen et al., [27] to enable chemomechanical systems. Finding a solution to convert fuels efficiently at small scales or in soft systems has the potential to revolutionize robotics and its applications, and should be studied extensively.
One of the most fruitful approaches to understanding and improving robotic performance has been comparison with biological systems. In the arena of bio-inspired robotic research, and even more in biomechanics, metrics of power and energy, normalized by mass or volume can be extremely useful for broad comparison and for identifying underlying similarities, as well as differences in locomotion. [28] The study of biological locomotion has focused on relative speed (body lengths s À1 ), cost of transport (distance Joule À1 ), and specific metabolic energy (J kg À1 ), to understand locomotion across the broad range of shapes and sizes of living systems. [29] For example, first-principles analysis of the energetic, force, and mass limitations on biological actuation have revealed that across 20 orders of magnitude in mass (from bacteria, to giraffes, to whales) the relative speed of movement is fairly constant. [30] Fundamentally, this behavior arises from energetic considerations of actuation, and holds true across a wide range of vastly different organisms. Similar "rules" of robotic locomotion may also exist across locomotion modality, actuator type, energy source, and power conversion mechanisms, and will be better understood through community focus on energy considerations for robotic performance.

Conclusion
Across these different platforms and technologies, researchers use bioinspiration to guide their work and find engineered solutions. In my research I often seek to compare the performance of sensors and actuators with natural counterparts at the same scale. By understanding relative performance to natural muscles, I develop an intuition for how suitable a specific technology is relative to the required task (e.g., locomotion, manipulation, etc.). As a community in robotics, we have forged collaborations across disciplines guided by the principles of bioinspiration and biomimicry. Recent events, such as the IROS 2021 workshop, on "Bio-inspired, Bio-mimetic and Bio-hybrid (Cyborg) Systems" show the strong interest across subfields of robotics in this arena. Bioinspiration is a great start, but we should be consistently evaluating against natural system performance by measuring the energy inputs and outputs of our robotic systems. By describing robots as energy systems and focusing on energy conversion throughout the robot's design and operation, we can better understand new technologies across scales and across disciplines.
Mihai Duduta is an assistant professor of mechanical engineering at the University of Toronto. For his doctoral thesis, Prof. Duduta developed a novel manufacturing technique for dielectric elastomer actuators (DEAs) as artificial muscles, which matched the speed and specific energy of natural muscles. He adapted the method to produce haptic devices, in collaboration with Facebook/Oculus. Afterward he won a Medical Devices Innovation Fellowship at the University of Minnesota, where he worked with engineers and clinicians to develop soft robotic tools for endovascular intervention. In Toronto, his research group studies electrically driven artificial muscles and soft rechargeable batteries.