Abstract
<jats:p>Micro aerial vehicles are emerging as powerful platforms for embodied autonomy in complex, GPS-denied, and hard-to-access environments. Their small size and agility allow them to operate under forest canopy, between tree lines, inside buildings, and in other cluttered settings where larger systems cannot go. At the same time, their severe constraints in sensing, computation, communication, energy, and payload make them a demanding testbed for physical AI: intelligence must be tightly coupled to perception, planning, control, and the dynamics of flight.</jats:p> <jats:p>In this talk, I will describe our work on intelligent micro aerial vehicles for autonomous operation in complex field environments. I will discuss the abstractions, architectures, and algorithms that enable robots to estimate state, reason about free space, plan dynamically feasible trajectories, and build semantic-metric maps in GPS-denied environments. I will also describe recent advances in learning-guided active perception, semantic mapping, air-ground collaboration, and language-specified missions, showing how MAVs can move beyond coverage and mapping toward task-driven information gathering.</jats:p> <jats:p>The talk will highlight applications in precision agriculture, forestry, infrastructure inspection, and search in unknown environments, with examples of autonomous flight under forest canopy, between tree rows, and in cluttered indoor and outdoor settings. I will close with a discussion of research needs for the next generation of MAVs, including scalable representations for physical AI, active perception with task-level value of information, safe real-time planning under uncertainty, onboard foundation models under size-weight-power constraints, resilient operation with degraded sensing and communication, and verification methods for field-deployed autonomy.</jats:p> <jats:p/>