Florian Berlinger, Melvin Gauci, and Radhika Nagpal. 2021. “Implicit coordination for 3D underwater collective behaviors in a fish-inspired robot swarm.” Science Robotics, 6, 50.Abstract
Many fish species gather by the thousands and swim in harmony with seemingly no effort. Large schools display a range of impressive collective behaviors, from simple shoaling to collective migration and from basic predator evasion to dynamic maneuvers such as bait balls and flash expansion. A wealth of experimental and theoretical work has shown that these complex three-dimensional (3D) behaviors can arise from visual observations of nearby neighbors, without explicit communication. By contrast, most underwater robot collectives rely on centralized, above-water, explicit communication and, as a result, exhibit limited coordination complexity. Here, we demonstrate 3D collective behaviors with a swarm of fish-inspired miniature underwater robots that use only implicit communication mediated through the production and sensing of blue light. We show that complex and dynamic 3D collective behaviors—synchrony, dispersion/aggregation, dynamic circle formation, and search-capture—can be achieved by sensing minimal, noisy impressions of neighbors, without any centralized intervention. Our results provide insights into the power of implicit coordination and are of interest for future underwater robots that display collective capabilities on par with fish schools for applications such as environmental monitoring and search in coral reefs and coastal environments.
Florian Berlinger, Mehdi Saadat, Hossein Haj-Hariri, George V Lauder, and Radhika Nagpal. 2021. “Fish-like three-dimensional swimming with an autonomous, multi-fin, and biomimetic robot.” Bioinspiration & Biomimetics, 16, 2. Publishers VersionAbstract
Fish migrate across considerable distances and exhibit remarkable agility to avoid predators and feed. Fish swimming performance and maneuverability remain unparalleled when compared to robotic systems, partly because previous work has focused on robots and flapping foil systems that are either big and complex, or tethered to external actuators and power sources. By contrast, we present a robot – the Finbot – that combines high degrees of autonomy, maneuverability, and biomimicry with miniature size (160 cm3). Thus, it is well-suited for controlled three-dimensional experiments on fish swimming in confined laboratory test beds. Finbot uses four independently controllable fins and sensory feedback for precise closed-loop underwater locomotion. Different caudal fins can be attached magnetically to reconfigure Finbot for swimming at top speed (122 mm/s ≡ 1 BL/s) or minimal cost of transport (CoT = 8.2) at Strouhal numbers as low as 0.53. We conducted more than 150 experiments with 12 different caudal fins to measure three key characteristics of swimming fish: (i) linear speed-frequency relationships, (ii) U-shaped costs of transport, and (iii) reverse Kármán wakes (visualized with particle image velocimetry). More fish-like wakes appeared where the cost of transport was low. By replicating autonomous multi-fin fish-like swimming, Finbot narrows the gap between fish and fish-like robots and can address open questions in aquatic locomotion, such as optimized propulsion for new fish robots, or the hydrodynamic principles governing the energy savings in fish schools.
Melinda Malley. 2020. “Army Ant Inspired Adaptive Self-Assembly with Soft Climbing Robots.” School of Engineering and Apllied Sciences (Mech. Eng), Harvard University. phdthesis2020malley.pdf
Julia Ebert, Melvin Gauci, Frederick Mallmann-Trenn, and Radhika Nagpal. 2020. “Bayes Bots: Collective Bayesian Decision-Making in Decentralized Robot Swarms.” Intl. Conference on Robotics and Automation (ICRA). icra2020-ebert.pdf
Melinda Malley, Bahar Haghighat, Lucie Houel, and Radhika Nagpal. 2020. “Eciton robotica: Design and Algorithms for an Adaptive Self-Assembling Soft Robot Collective.” Intl. Conferenc on Robotics and Automation (ICRA). icra2020-malley.pdf
Paul Bardunias, Daniel Calovi, Nicole Carey, Rupert Soar, J. Scott Turner, Radhika Nagpal, and Justin Werfel. 2020. “The extension of internal humidity levels beyond the soil surface facilitates mound expansion in Macrotermes.” Proceedings of the Royal Society B: Biological Science, 287, 1930.Abstract
Termites in the genus Macrotermes construct large-scale soil mounds above their nests. The classic explanation for how termites coordinate their labour to build the mound, based on a putative cement pheromone, has recently been called into question. Here, we present evidence for an alternate interpretation based on sensing humidity. The high humidity characteristic of the mound's internal environment extends a short distance into the low-humidity external world, in a ‘bubble’ that can be disrupted by external factors like wind. Termites transport more soil mass into on-mound reservoirs when shielded from water loss through evaporation, and into experimental arenas when relative humidity is held at a high value. These results suggest that the interface between internal and external conditions may serve as a template for mound expansion, with workers moving freely within a zone of high humidity and depositing soil at its edge. Such deposition of additional moist soil will increase local humidity, in a feedback loop allowing the ‘interior’ zone to progress further outward and lead to mound expansion.
Mihai Duduta, Florian Berlinger, Radhika Nagpal, David Clarke, Rob Wood, and Zeynep Temel. 2020. “Tunable Multi-Modal Locomotion in Soft Dielectric Elastomer Robots.” IEEE Robotics and Automation Letters (RAL). ral2020_duduta.pdf
Daniel Calovi, Paul Bardunias, Nicole Carey, J. Scott Turner, Radhika Nagpal, and Justin Werfel. 2019. “Surface curvature guides early construction activity in mound-building termites.” Philosophical Transactions of the Royal Society , 374, 1774. Publisher's VersionAbstract

Termite colonies construct towering, complex mounds, in a classic example of distributed agents coordinating their activity via interaction with a shared environment. The traditional explanation for how this coordination occurs focuses on the idea of a ‘cement pheromone’, a chemical signal left with deposited soil that triggers further deposition. Recent research has called this idea into question, pointing to a more complicated behavioural response to cues perceived with multiple senses. In this work, we explored the role of topological cues in affecting early construction activity in Macrotermes. We created artificial surfaces with a known range of curvatures, coated them with nest soil, placed groups of major workers on them and evaluated soil displacement as a function of location at the end of 1 h. Each point on the surface has a given curvature, inclination and absolute height; to disambiguate these factors, we conducted experiments with the surface in different orientations. Soil displacement activity is consistently correlated with surface curvature, and not with inclination nor height. Early exploration activity is also correlated with curvature, to a lesser degree. Topographical cues provide a long-term physical memory of building activity in a manner that ephemeral pheromone labelling cannot. Elucidating the roles of these and other cues for group coordination may help provide organizing principles for swarm robotics and other artificial systems.

This article is part of the theme issue ‘Liquid brains, solid brains: How distributed cognitive architectures process information’.

Helen McCreery, Jenna Bilek, Radhika Nagpal, and Michael Breed. 2019. “Effects of load mass and size on cooperative transport in ants over multiple transport challenges.” Journal of Experimental Biology, doi: 10.1242/jeb.206821. Publisher's Version (open access)Abstract
Some ant species cooperatively transport a wide range of extremely large, heavy food objects of various shapes and materials. While previous studies have examined how object mass and size affect the recruitment of additional workers, less is understood about how these attributes affect the rest of the transport process. Using artificial baits with independently varying mass and size, we reveal their effects on cooperative transport in Paratrechina longicornis across two transport challenges: movement initiation and obstacle navigation. As expected, object mass was tightly correlated with number of porters as workers adjust group size to the task. Mass affected performance similarly across the two challenges, with groups carrying heavy objects having lower performance. Yet, object size had differing effects depending on the challenge. While larger objects led to reduced performance during movement initiation – groups took longer to start moving these objects and had lower velocities – there was no evidence for this during obstacle navigation, and the opposite pattern was weakly supported. If a group struggles to start moving an object, it does not necessarily predict difficulty navigating around obstacles; groups should persist in trying to move ‘difficult’ objects, which may be easier to transport later in the process. Additionally, groups hitting obstacles were not substantially disrupted, and started moving again sooner than at the start, despite the nest direction being blocked. Paratrechina longicornis transport groups never failed, performing well at both challenges while carrying widely varying objects, and even transported a bait weighing 1900 times the mass of an individual.
Mihai Duduta, Florian Berlinger, Radhika Nagpal, David Clarke, Robert Wood, and Zeynep Temel. 2019. “Electrically-latched compliant jumping mechanism based on a dielectric elastomer actuator.” Smart Materials and Structures , 28, 09LT01. duduta_sms_letter_2019.pdf
Lucie Houel. 2019. “Self-assembly of soft-robots in simulation inspired by army ant bridge behavior.” EPFL Master's Thesis.
Katherine Binney. 2019. “Teach a Fish to Swim: Evaluating the Ability of Turing Learning to Infer Schooling Behavior.” Senior Thesis, Harvard University.Abstract

Turing Learning is a promising evolutionary design method for swarm robotics that uses ob- servation of natural or artificial systems to infer controllers for agents in a swarm. However, Tur- ing Learning has thus far only been used to infer very simple swarm behaviors. In this work, we expand Turing Learning to infer dispersion, a much more complex swarm behavior, by a simulated school of robotic fish. Turing Learning depends on the co-evolution of replicas and classifiers. Replicas mimic ideal behavior and classifiers distinguish between data samples from replica and ideal agents. We model replicas and classifiers with neural networks and investigate the architecture of each component independently in order to determine needed modifications to Turing Learning for it to infer fish schooling. We find that previously formulated data samples led to the inference of behaviors that locally mimicked the agent trajectories in dispersion, yet poorly mimicked dispersion of an entire swarm. We present three alternative data samples that consider the spatial arrangement of agents in a swarm. We also introduce three new classifier fitness func- tions that accelerate evolution of high-accuracy classifiers. We find in a preliminary trial that using one of our data samples (metrics) and classifier fitness functions (foutputs) enables the successful inference of dispersion via Turing Learning.

Katherina Soltan, Jamie O'Brien, Florian Berlinger, Radhika Nagpal, and Jeff Dusek. 2018. “A Biomimetic Actuation Method for a Miniature, Low-Cost Multi-jointed Robotic Fish.” In IEEE MTS OCEANS. oceans2018soltan.pdf
Florian Berlinger, Mihai Duduta, Hudson Gloria, David Clarke, Radhika Nagpal, and Robert Wood. 2018. “A Modular Dielectric Elastomer Actuator to Drive Miniature Autonomous Underwater Vehicles.” In Intl. Conf. on Robotics and Automation (ICRA).Abstract

Abstract—In this paper we present the design of a fin-like dielectric elastomer actuator (DEA) that drives a miniature autonomous underwater vehicle (AUV). The fin-like actuator is modular and independent of the body of the AUV. All electronics required to run the actuator are inside the 100 mm long 3D-printed body, allowing for autonomous mobility of the AUV. The DEA is easy to manufacture, requires no pre-stretch of the elastomers, and is completely sealed for underwater operation. The output thrust force can be tuned by stacking multiple actuation layers and modifying the Young’s modulus of the elastomers. The AUV is reconfigurable by a shift of its center of mass, such that both planar and vertical swimming can be demonstrated on a single vehicle. For the DEA we measured thrust force and swimming speed for various actuator designs ran at frequencies from 1Hz to 5Hz. For the AUV we demonstrated autonomous planar swimming and closed- loop vertical diving. The actuators capable of outputting the highest thrust forces can power the AUV to swim at speeds of up to 0.55body lengths per second. The speed falls in the upper range of untethered swimming robots powered by soft actuators. Our tunable DEAs also demonstrate the potential to mimic the undulatory motions of fish fins. 

(finalist for Best Conference Paper and Best Student Paper Awards)
Julia Ebert, Melvin Gauci, and Radhika Nagpal. 2018. “Multi-Feature Collective Decision Making in Robot Swarms.” In Intl. Conf. on Autonomous Agents and Multiagent Systems (AAMAS).Abstract

Collective decision making has been studied extensively in the fields of multi-agent systems and swarm robotics, inspired by its pervasiveness in biological systems such as honeybee and ant colonies. However, most previous research has focused on collective decision making on a single feature. In this work, we introduce and investigate the multi-feature collective decision making problem, where a collective must decide on multiple binary features simultaneously, given no a priori information about their relative difficulties. Each agent may only estimate one feature at any given time, but the agents can locally communicate their noisy estimates to arrive at a decision. We demonstrate a decentralized algorithm for single-feature decision making and a dynamic task allocation strategy that allows the agents to lock in decisions on multiple features in finite time. We validate our approach using simulated and physical Kilobot robots. Our results show that a collective can correctly classify a multi-feature environment, even if presented with pathological initial agent-to-feature allocations. 

Melinda Malley, Michael Rubenstein, and Radhika Nagpal. 2017. “Flippy: A Soft, Autonomous Climber with Simple Sensing and Control.” In IEEE/RSJ Intl Conference on Intelligent Robots and Systems (IROS).Abstract

Climbing robots have many potential applications including maintenance, monitoring, search and rescue, and self-assembly. While numerous climbing designs have been investigated, most are limited to stiff components. Flippy (Fig. 1) is a small, flipping biped robot with a soft, flexible body and on-board power and control. Due to its built-in compliance, flipping gait, and corkscrew gripper, it can autonomously climb up and down surfaces held at any angle relative to gravity and transition from one surface to another, without complex sensing or control. In this paper, we demonstrate the robot’s ability to flip consistently over a flat Velcro surface and 2D Velcro track, where it reliably climbs vertically, upside down and back to a flat surface, completing all the interior transitions in-between. 

Florian Berlinger, Jeff Dusek, Melvin Gauci, and Radhika Nagpal. 2017. “Robust Maneuverability of a Miniature Low-Cost Underwater Robot using Multiple Fin Actuation.” IEEE Robotics and Automation Letters (RA-L), PP, 99.Abstract

In this paper we present the design of a miniature (100 mm) autonomous underwater robot that is low-cost ($ 100), easy to manufacture, and highly maneuverable. A key aspect of the robot design that makes this possible is the use of low-cost magnet-in-coil actuators, which have a small profile and minimal sealing requirements. This allows us to create a robot with multiple flapping fin propulsors that independently control robot motions in surge, heave, and yaw. We present several results on the robot, including: (i) quantified open-loop swimming characteristics; (ii) autonomous behaviors using a pressure sensor and an IMU to achieve controlled swimming of prescribed trajectories; (iii) feedback from an optic sensor to enable homing to a light source. The robot is designed to form the basis for underwater swarm robotics testbeds, where low cost and ease of manufacture are critical, and 3D maneuverability allows testing complex coordination inspired by natural fish schools. Individually, miniature and low-cost underwater robots can also provide a platform for the study of 3D autonomy and marine vehicle dynamics in educational and resource-constrained settings. 

Kirstin Petersen and Radhika Nagpal. 2017. “Complex Design by Simple Robots: A Collective Embodied Intelligence Approach to Construction.” Architectural Design, Special Issue: Autonomous Assembly: Designing for a New Era of Collective Construction, 87, 4, Pp. 44-49.Abstract
Nature's builders – from termites to beavers – offer a model of collective intelligence that can inspire robotic construction. Kirstin Petersen, Assistant Professor in Electrical and Computer Engineering at Cornell University, Ithaca, New York, and Radhika Nagpal, Professor in Computer Science at the Harvard School of Engineering and Applied Sciences, Cambridge, Massachusetts, describe several recent projects in this field that they have been involved in, both separately and collaboratively.
Tsvetomira (Mira) Radeva, Anna Dornhaus, Nancy Lynch, Radhika Nagpal, and Hsin-Hao Su. 2017. “Costs of task allocation with local feedback: Effects of colony size and extra workers in social insects and other multi-agent systems.” PLoS computational biology, 13, 12. Publisher's Version (open-access)Abstract
(Author summary) 

Many complex systems have to allocate their units to different functions: cells in an embryo develop into different tissues, servers in a computer cluster perform different cal- culations, and insect workers choose particular tasks, such as brood care or foraging. Here we demonstrate that this process does not automatically become easier or harder with sys- tem size. If more workers are present than needed to complete the work available, some workers will always be idle; despite this, having surplus workers makes redistributing them across the tasks that need work much faster. Thus, unexpectedly, such surplus, idle workers may potentially significantly improve system performance. Our work suggests that interdisciplinary studies between biology and distributed computing can yield novel insights for both fields. 

Melvin Gauci, Monica Ortiz, Michael Rubenstein, and Radhika Nagpal. 2017. “Error Cascades in Collective Behavior: A Case Study of the Gradient Algorithm on 1000 Physical Agents.” In 16th International Conference on Autonomous Agents and Multiagent Systems (AAMAS).Abstract

The gradient, or hop count, algorithm is inspired by nat- ural phenomena such as the morphogen gradients present in multi-cellular development. It has several applications in multi-agent systems and sensor networks, serving as a basis for self-organized coordinate system formation, and finding shortest paths for message passing. It is a simple and well- understood algorithm in theory. However, we show here that in practice, it is highly sensitive to specific rare errors that emerge at larger scales. We implement it on a system of 1000 physical agents (Kilobot robots) that communicate asynchronously via a noisy wireless channel. We observe that spontaneous, short-lasting rare errors made by a sin- gle agent (e.g. due to message corruption) propagate spa- tially and temporally, causing cascades that severely hinder the algorithm’s functionality. We develop a mathematical model for temporal error propagation and validate it with experiments on 100 agents. This work shows how multi- agent algorithms that are believed to be simple and robust from theoretical insight may be highly challenging to im- plement on physical systems. Systematically understanding and quantifying their current limitations is a first step in the direction of improving their robustness for implementation.