Brair (Tilboon) Elberier

Vision-based autonomy can substantially expand the capabilities of insect-scale mobile robots, but integrating visual perception and control under strict size, weight, and power (SWaP) constraints remains challenging at the near-gram scale. At this size, camera capture, onboard inference, wireless communication, and motor control must share severely constrained computational and memory resources while maintaining perception rates sufficient for closed-loop behavior. We present MilliMobile Vision, an extensible mobile robotic platform that integrates onboard visual perception, inference, and closed-loop control within a system weighing less than 4 g and measuring 2×2×2 cm. The platform uses off-the-shelf components and an extensible hardware architecture with exposed interfaces for integrating additional sensors and peripherals, while its wheel-based locomotion enables speeds up to 70 cm/s. We develop a compute-aware perception–control architecture that separates lightweight visual detection and tracking from computationally intensive learned classification. Using only a single monochrome camera for environmental perception, lightweight image processing provides target localization and obstacle awareness for closed-loop navigation at approximately 3 Hz, while a 27-class neural network is selectively invoked for onboard insect identification with approximately 80% validation accuracy. Together, these components enable an untethered closed-loop task in which the robot autonomously searches for, identifies, approaches, and follows small moving targets using only onboard visual feedback and computation. This hierarchical perception strategy reserves learned inference for observations where semantic information is required, preserving visual feedback for responsive locomotion under constrained onboard compute and memory. We characterize the resulting perception latency, inference, and closed-loop control trade-offs and demonstrate sustained autonomous tracking of continuously moving targets, including a freely moving live insect.

The proposed computer vision (CV) informed parameter estimation pipeline is capable of generating parameter estimates for a selected linear dynamic system. This is accomplished by (1) collecting a time-series dataset of a system's physical motion, (2) relating the captured motion to a designed physics-based model, and (3) estimating the most probable set of parameters that represents the observed motion using Bayesian inference. This analysis provides a fundamental understanding of the system's intrinsic properties that enables the refinement and generation of new system affordances. Leveraging this additional knowledge, external learning systems may use this physics-based approach to make predictions of a system's potential environmental interactions. We compare the results of our generalized approach with those of a handcrafted solution designed for a similarly constructed experiment. Our experiments reveal a MAPE of 3.962%, highlighting the approach's robustness given a time-series dataset subject to additive white Gaussian noise (AWGN).

Using wind to disperse microfliers that fall like seeds and leaves can help automate large-scale sensor deployments. Here, we present battery-free microfliers that can change shape in mid-air to vary their dispersal distance. We designed origami microfliers using bistable leaf-out structures and uncovered an important property: a simple change in the shape of these origami structures causes two dramatically different falling behaviors. When unfolded and flat, the microfliers exhibit a tumbling behavior that increases lateral displacement in the wind. When folded inward, their orientation is stabilized, resulting in a downward descent that is less influenced by wind. To electronically transition between these two shapes, we designed a low-power electromagnetic actuator that produces peak forces of up to 200 millinewtons within 25 milliseconds while powered by solar cells. We fabricated a circuit directly on the folded origami structure that includes a programmable microcontroller, a Bluetooth radio, a solar power–harvesting circuit, a pressure sensor to estimate altitude, and a temperature sensor. Outdoor evaluations show that our 414-milligram origami microfliers were able to electronically change their shape mid-air, travel up to 98 meters in a light breeze, and wirelessly transmit data via Bluetooth up to 60 meters away, using only power collected from the sun.