Mechanical and Civil Engineering Seminar: PhD Thesis Defense

Abstract:

Achieving agile, efficient, and robust locomotion in bipedal robots remains a grand challenge of robotics.Traditional model-based control methods are theoretically grounded but are often sensitive to model mismatch andstate-estimation uncertainty, limiting their adaptability to real-world environments. Conversely, data-driven approachessuch as reinforcement learning produce remarkable behaviors but often lack interpretability, require non-trivial rewardshaping, and raise safety concerns.This thesis bridges these two paradigms through a unified framework that begins with model-based behavior synthesisand culminates in data-driven adaptation. The first part focuses on constructing walking behaviors and controllers usingreduced-order models of locomotion. A hierarchy of planners and controllers is developed to enable robust walking forflat-footed and multi-domain gaits, as well as safety-critical locomotion over constrained footholds such as stairs andstepping stones. Additionally, this work introduces RoMoCo, a modular open-source architecture, a modular open-source architecture designed to unify reduced-order planning, output synthesis, and whole-body control across multiplebipedal platforms.Building on this foundation, the second part introduces data-driven mechanisms that enable robots to improve andpersonalize their behaviors through various forms of data. Episodic data collected during repeated executions are used tocorrect modeling errors and reduce constraint violations. Human preference data facilitates automatic gain tuningthrough interactive feedback. Online robot data enables adaptation of reduced-order models by learning step-to-stepdynamics directly from real executions. Environmental interaction data inform high-level planners for navigatinginfeasible paths. Finally, large-scale simulation data support a reinforcement-learning framework designed for hardwaredeployment, where model-guided rewards enable efficient training and introduce perception inputs, yielding policiescapable of dynamic stepping-stone traversal on real robots.Together, these contributions form a progression from theoretically grounded model-based control to data-enabledadaptation, demonstrating that reduced-order models and data-driven learning are complementary. Their integrationenables bipedal robots such as Cassie and G1 to walk safely, robustly, and efficiently across diverse terrains, marking astep toward human-level agility in legged locomotion.