Jake Welde – Geometric Abstractions for Efficient and Explainable Control of Complex Aerial Robots 

Abstract:

Aerial robots have the potential to move dynamically through unsafe, cluttered, or hard-to-reach environments to perform vital tasks that humans cannot. However, to achieve the morphological complexity necessary for physical interaction, today’s aerial robots sacrifice dynamic behavior—only simple, single-body vehicles like quadrotors fly acrobatically, whereas bulky, complex systems move sluggishly and cautiously. On the contrary, complex biological organisms like hummingbirds demonstrate incredible dexterity and agility simultaneously, far outstripping current robotic systems. To realize even a fraction of these capabilities, I believe we must jointly explore the combined control-morphology design space of robotic systems.

In particular, I argue that differential geometry offers a remarkably effective toolkit for developing efficient control algorithms that also inform morphology design. By leveraging the natural Lie group symmetries underlying the mechanics, we enable efficient planning of dynamically feasible trajectories for underactuated systems and accelerate reinforcement learning for trajectory tracking control with improved generalization. Such control insights also guide design, closing the control-morphology feedback loop and leading to synergies between a robot’s embodiment and its controller. By combining explainable abstractions with scalable computation, I build towards a future in which aerial robots interact with their surroundings as dynamically and capably as their counterparts in Nature.

Bio:

Jake Welde is a PhD candidate in Mechanical Engineering and Applied Mechanics at the University of Pennsylvania in the General Robotics, Automation, Sensing, and Perception (GRASP) Laboratory, working with Vijay Kumar. He explores the role of differential geometry and dynamical systems theory in control synthesis and design for robotic systems, using these tools to explainably synthesize explicit controllers, accelerate learning algorithms, and develop more capable robot morphologies. Jake is the recipient of the NSF GRFP ‘19 and a member of the RSS Pioneers ‘24 cohort, and his interdisciplinary research earned a Best Paper Award at NeurReps ‘24 and a Best Paper Finalist mention at ICRA ‘21. His contributions as an educator and departmental community member have been recognized with the Outstanding Teaching Assistant Award and the John A. Goff Prize for scholarship, resourcefulness, and leadership.

Details

• Date: Monday, March 17
• Time: 10:00 am - 11:00 am
• Category: Special Seminar
• Location: Grier A (34-401A)

Host

• Jeffrey Lang
• Email: chadcoll@mit.edu