School of Aerospace Engineering

Gebhardt Distinguished Lecture 

 

 

Resilient Autonomy: Perception and Planning 
for Dynamic, Unknown Environments

 

featuring

 

Jonathan P. How 

Ford Professor of Engineering | Massachusetts Institute of Technology

 

Thursday, November 13
11am - 12pm
Guggenheim 442

 About the Seminar: 
Unmanned aerial systems (UAS) hold promise for critical applications such as search and rescue, environmental monitoring, and autonomous delivery. However, deploying them in real-world, safety-critical settings presents core challenges: navigating GPS-denied environments, reasoning under uncertainty, and planning safe trajectories in dynamic, partially known spaces. This talk presents recent advances in perception and planning that together enable robust, scalable, and efficient aerial autonomy. On the perception side, we introduce several complementary mapping frameworks. GRANDSLAM fuses 3D Gaussian splatting with semantics and geometric priors to create unified scene representations for photorealistic planning. ROMAN compresses environments into sparse, object-centric maps that are orders of magnitude smaller than traditional representations while still supporting accurate relocalization and loop closure under extreme viewpoint changes. On the planning side, we first introduce IL-RTMPC, a demonstration- and training-efficient method for learning robust control policies from model predictive control (MPC). It combines single-trajectory demonstrations with a disturbance-aware data aggregation strategy to produce policies that generalize to unseen conditions. We demonstrate its effectiveness on quadrotors and the MIT SoftFly. DYNUS then provides uncertainty-aware trajectory planning for safe, real-time flight in dynamic, unknown environments. Building on this, MIGHTY performs fully coupled spatiotemporal optimization, generating agile, precise motion by jointly reasoning about path and timing. Combined with prior work on Robust-MADER, these methods deliver compact, consistent maps and robust trajectory optimization that support fast, safe multi-robot navigation in complex environments. I will present experimental results across simulation and hardware platforms and conclude with open challenges in building resilient, real-world UAS autonomy. These advances bring us closer to reliable autonomous aerial systems with meaningful impact in real-world operations. About the Speaker: Jonathan P. How is the Ford Professor of Engineering at the Massachusetts Institute of Technology. He received his S.M. and Ph.D. in Aeronautics and Astronautics from MIT and was previously on the faculty at Stanford before joining MIT in 2000. His research focuses on robust planning and learning under uncertainty, with emphasis on multi-agent systems and autonomous flight. He is a Fellow of IEEE and AIAA, was elected to the National Academy of Engineering in 2021, and his work has been recognized with numerous awards including the IEEE Transactions on Robotics King-Sun Fu Memorial Best Paper Awards in 2022 and 2024.