Yiannis Aloimonos
University of Maryland
Meet the invited speakers for the Insect-scale Autonomy Workshop at IROS 2026.
Invited Speakers
Sean Humbert
University of Colorado Boulder
Frances Chance
Sandia National Laboratories
Sarah Bergbreiter
Carnegie Mellon University
Sawyer Fuller
University of Washington
Geoffrey Barrows
Centeye, Inc.
Elia Cereda
Dalle Molle Institute for Artificial Intelligence
Gabriel Gattaux
Aix-Marseille University and CNRS
Invited Talks
Explore the talk abstracts and speaker biographies below.
Speaker
Yiannis Aloimonos
University of Maryland
Keynote
Purposive Memory and Purposive Perception: A key to insect-scale autonomy
Abstract
Robots are built to act in the real world; at the insect scale, power, memory, sensing, and computation become severe constraints. We propose a different foundation for autonomy: a purposive perception–memory–action loop. Our central thesis is that navigation is not primarily a mapping problem; it is a memory problem, shaped by what perception chooses to extract for action. Insects provide a biological proof of principle. Bees, ants, and flies navigate complex environments with remarkably small nervous systems by extracting behaviorally relevant sensory information and maintaining compact navigational variables. What we take from insects is a computational principle, not a blueprint: perceive only what updates the task state, and remember only what must persist for future action. For robotics, this means memory should match the structure of the variable it represents: a ring for heading, and richer structures only when the behavior demands them. Through experiments in purposive perception and structured spatial memory, we show how such frugal representations keep a robot acting under degraded and intermittent sensing, showing how frugality and structure can become sources of robustness. The guiding question is not “how do I reconstruct the world?” but “what is the least a robot must perceive and remember in order to act?”
A Socratic Dialog – Yiannis Aloimonos and Naitri Rajyaguru, Computer Vision Lab, Univ. of Maryland
Speaker Bio
Yiannis Aloimonos is Professor of Computational Vision and Intelligence at the Department of Computer Science, University of Maryland, College Park, and the Director of the Computer Vision Laboratory at the Institute for Advanced Computer Studies (UMIACS). He is also affiliated with the Institute for Systems Research and the Neural and Cognitive Science Program. He was born in Sparta, Greece and studied Mathematics in Athens and Computer Science at the University of Rochester, NY (PhD 1990). He is interested in Active Perception and the modeling of vision as an active, dynamic process for real time robotic systems. For the past five years he has been working on bridging signals and symbols, specifically on the relationship of vision to reasoning, action and language. He received the Presidential Young Investigator Award from President Bush and the Bodossaki Prize in Artificial Intelligence. He is an IEEE Fellow.
Speaker
Sean Humbert
University of Colorado Boulder
Keynote
Sensing What Matters: Embodied Control for Agile Autonomous Systems
Abstract
Our team’s field experience in the DARPA Subterranean Challenge revealed a central bottleneck in autonomous exploration: physical agility and robustness in uncertain, cluttered environments. In contrast, natural organisms negotiate complex terrain with remarkable ease—a lizard scrambling across rocks or a bird swerving through dense vegetation. Why can’t our robotic systems match this performance? Is the advantage better hardware, or a fundamentally different sensing and control architecture? This talk explores a shift from model-centric autonomy toward proprioceptive sensing and control inspired by biology. Rather than relying on increasingly detailed dynamic models, we leverage the structure of the platform itself as a sensing substrate. Distributed mechanosensing embedded in the airframe provides low-latency access to dynamically significant quantities without numerical differentiation or heavy filtering. By grounding control in direct physical observability rather than complex modeling, we move closer to the principles underlying biological agility in uncertain environments.
Speaker Bio
Dr. Sean Humbert is the Denver Business Challenge Professor in the Department of Mechanical Engineering and the founding Director of the Robotics Graduate Program at the University of Colorado Boulder. He holds a BS degree in Mechanical Engineering from the University of California Davis along with MS and PhD degrees in Mechanical Engineering from Caltech. As PI of the Bio-Inspired Perception and Robotics Laboratory, his group applies control- and information-theoretic tools to distill perception and reduction principles in small animals such as insects, providing insight into the biology and resulting in novel, robust, and computationally efficient solutions for engineered systems. Prof. Humbert has served as a member of the Board for Army Science and Technology (BAST), as a core member of the Board on Army RDT&E, Systems Acquisition and Logistics (BARSL), as a member of the DARPA MTO Microsystems Exploratory Council (MEC), and is an AIAA Associate Fellow. He is the recipient of the AIAA National Capital Section Hal Andrews Young Scientist/Engineer Award and an ARO Young Investigator Award.
Speaker
Frances Chance
Sandia National Laboratories
Keynote
The Computer Bug You Want: Insect-Inspired Neuromorphic Primitives for Energy-Efficient Computation
Abstract
Animals excel at a range of essential behaviors, for example hunting or foraging, that require fast calculations under tight energy constraints. Identifying key computational primitives of biological nervous systems is critical for developing artificial systems that can similarly perform even with limited energy budgets. Insects perform surprisingly complex tasks given the relatively small size of their brains. I will present work focused on developing energy-efficient neuromorphic computing primitives derived from single-neuron computational primitives of insect nervous systems. For example, we have recently developed an analog neuromorphic emulation of shunting inhibition, a biophysical mechanism that approximates real-time multiplication in the Drosophila T4 dendrite, and leveraged this circuit primitive in a range of biologically-inspired applications, including coordinate transformations and dynamic gain normalization. I will discuss ongoing work to leverage these neuromorphic primitives for energy-efficient artificial intelligence models, and how these models may impact on-board compute and edge-AI hardware.
Speaker Bio
Frances Chance received her MS and PhD from Brandeis University, and her BS from the California Institute of Technology. She is currently a Distinguished Member of the Technical Staff in the Center for Computing Research at Sandia National Laboratories. Her research applies knowledge of biological nervous systems and neural circuit operations to develop and constrain novel neural-informed algorithms and brain-based technologies.
Speaker
Sarah Bergbreiter
Carnegie Mellon University
Keynote
Event-based mechanosensing for insect-scale autonomy
Abstract
Insects use neural mechanosensing to process information efficiently for agile, robust flight. We explore two types of mechanosensors for disturbance detection in flight: strain and flow. Inspired by neurons embedded in insect wings that efficiently encode complex strain patterns through nonlinear filtering, we designed and fabricated strain-sensitive "switches" directly on flexible wings. These switches open and close at a designated strain threshold, and the timing of their closures allows us to detect and classify disturbances and body rotations (sinusoidal yaw, 4 rad/s amplitude). We also explore sensor placement on the wing computationally and experimentally. Motivated by the many hairs on insect wings and bodies, we extended the same switch-based approach to flow sensing: sub-millimeter sensors respond to airflow reversal (useful for stall detection) and velocities as low as 0.5 m/s.
Speaker Bio
Sarah Bergbreiter is the Dan and Karen Swanson Endowed Professor of Mechanical Engineering at Carnegie Mellon University, where she also serves as Associate Head for Strategic Initiatives. Her research integrates robotics, MEMS, materials, and manufacturing to create new sensing, actuation, and locomotion technologies for small-scale robotic systems. Prof. Bergbreiter's honors include the DARPA Young Faculty Award, NSF CAREER Award, and Presidential Early Career Award for Scientists and Engineers (PECASE). She and her fabulous current and former students have also received several Best Paper awards at ICRA, IROS, and the Hilton Head Workshop. She is a Fellow of ASME and previously served as Vice Chair of DARPA's Microsystems Exploratory Council. Outside of academia, she enjoys spending time with her husband and two daughters, running or biking outside rather slowly, and the rare game of water polo.
Speaker
Sawyer Fuller
University of Washington
Keynote
Embodied Intelligence in Insect Robotics
Abstract
The theory of embodied intelligence holds that the way brains think is inexorably tied to the body. Some portion of that intelligence comes from the mechanics of the body itself. Size reductions made possible by advances in microfabrication—from sensors to actuators to mechanisms—will allow future autonomous systems to reduce in size to that of an insect. But like in biology, insect-sized robots will never be able to match their larger counterparts in intelligence: there is no room for big computers or batteries. My research team creates robots less than a gram that address this challenge by imbuing their bodies with smart physical design to complement simple computation. Like insects, despite their small size, they can sense and respond to their environment. Flying and hopping robots at this scale could act as mobile sensors, helping detect gas leaks, spot early forest fires, monitor crops, or track the spread of airborne diseases.
Speaker Bio
Sawyer Fuller creates biologically-inspired sensors, control systems, and mechanical designs targeted at insect-sized air and ground vehicles, and investigates the flight systems of aerial insects. He completed his Ph.D. in Biological Engineering at the California Institute of Technology and B.S. and M.S. degrees in Mechanical Engineering at the Massachusetts Institute of Technology, and postdoctoral training at Harvard. In addition to his work in insect flight control, he also developed a frog-hopping robot at the NASA Jet Propulsion Laboratory and invented an ink-jet printer capable of fabricating millimeter-scale 3D metal machines at the MIT Media Lab. His work at the intersection of robotics and biology has appeared in journals such as Science and Proceedings of the National Academy of Sciences.
Speaker
Geoffrey Barrows
Centeye, Inc.
Industry Talk
What can you do with a few thousand pixels?
Abstract
The costliest element in the SWaP budget of a small robotic vision system is generally not the image sensor but the processing behind it. In earlier work we implemented several vision-based flight control tasks using just tens to hundreds of pixels. Reflections on what worked, and on the differences between contemporary computer vision and biological vision, led us to a metric we call pixels per frame: the visual motion, measured in pixels or photoreceptors, accruing in one frame or update. Conventional computer vision operates above one pixel per frame, which carries steep downstream processing costs. Our implementations operate below this threshold, as do event cameras, and we argue biological systems do as well. Below that line, new kinds of image sensor arrays and new classes of algorithms become available, and the processing burden drops sharply. This talk covers the metric, example image sensors, and use cases.
Speaker Bio
Geoffrey Barrows is the founder of Centeye, Inc., where he develops neuromorphic vision chips and low-SWaP sensing systems for difficult applications. He holds a PhD from the University of Maryland, College Park, and previously worked at the Naval Research Laboratory. His interest in insect-inspired vision dates to 1999, when he built and flew a neuromorphic optical flow sensor on a small air platform.
Speaker
Elia Cereda
Dalle Molle Institute for Artificial Intelligence
Rising Star Talk
Closing the Loop at Insect Scale: Lessons from Performance-optimized Software Architectures for Nano-UAVs
Abstract
Insect-scale robots must rely on ultra-low power resource-constrained MCUs for perception-to-action onboard intelligence. At the nano-UAV scale, i.e., 10 cm, tens of grams, and sub-100mW perception, much research concentrated on TinyML models, controllers, and bio-inspired algorithms, while treating the underlying software infrastructure as an implementation detail. This choice is costly: across state-of-the-art nano-UAV systems, closed-loop throughput falls 16–92% short of inference workloads in isolation. Our in- field experiments show the lost throughput directly degrades closed-loop performance, e.g., up to 30% higher position error and mission success dropping from 100% to 40%. This talk builds on NanoCockpit, our open-source performance-optimized application framework for the 27-gram Crazyflie that recovers the ideal throughput, now being integrated into the platform's official software. From this experience, the talk distills broader lessons for insect-scale autonomy, such as zero-copy, memory-efficient software abstractions, latency-efficient sensing-to-control pipelines, and the software stack as a pillar of the overall system design.
Speaker Bio
Elia Cereda is a postdoctoral researcher at the Dalle Molle Institute for Artificial Intelligence (IDSIA, USI-SUPSI) in Lugano, Switzerland, where he received his Ph.D. in 2026. He researches robust AI for autonomous pocket-sized robotic platforms, spanning performance-optimized software infrastructure, TinyML generalization under domain shift, and self-supervised on-device learning to bridge the gap between lab and real world. He authored more than 15 publications, including Best Paper Awards at IEEE ICCE'18 and ACM EWSN'23 SPICES, and the first demonstrations of on-device learning aboard nano-drones. He was part of the winning team of the first "Nanocopter AI Challenge" at IMAV'22.
Speaker
Gabriel Gattaux
Aix-Marseille University and CNRS
Rising Star Talk
How Insect Brains Inspire Frugal Autonomous Navigation
Abstract
Solitary foraging ants navigate complex environments with remarkably limited sensory and neural resources, while autonomous robots often rely on power-hungry sensors, large memories, and intensive computation. This talk explores how insect vision, neural mechanisms, and behavior can inspire frugal and robust robotic navigation. I will present bio-inspired models of visual memory and decision-making implemented on ground and aerial robots. Through visual homing and route following, I will show how low-resolution vision, compact neural representations, and active scanning enable reliable wayfinding with minimal sensing, memory, and computation. Beyond engineering applications, these systems provide embodied tools for investigating insect navigation and raise broader questions about the possible functions of specific brain regions and neuronal populations. This reciprocal approach illustrates how studying tiny brains can advance autonomous robotics while deepening our understanding of biological intelligence, particularly by asking whether complex navigational behaviors require a cognitive map or might emerge from simpler, compass-like representations.
Speaker Bio
Gabriel Gattaux is completing his PhD in biorobotics at Aix-Marseille University (AMU-CNRS). His interdisciplinary training spans mechanical engineering at the University of Lorraine, mechatronics at ENSIL-ENSCI, University of Limoges, and computer science at Wrocław University of Science and Technology. Bridging robotics, vision, control theory, computational neuroscience, ethology, and neuromorphic engineering, his research investigates how insects perceive, navigate, and make decisions, particularly during wayfinding. He translates these biological insights into frugal, robust robotic navigation systems that operate with minimal sensing, memory, and computation. His recent work has been published in Nature Communications and IEEE Robotics and Automation Letters.