///Optimal control of compliant bipedal gaits and their implementation on robot hardware
Loading Events
This event has passed.

PhD Thesis Defense

February

15
Fri
William Martin PhD Student Robotics Institute,
Carnegie Mellon University
Friday, February 15
8:30 am
- 9:30 am
NSH 3305
Optimal control of compliant bipedal gaits and their implementation on robot hardware

Abstract:
Legged animals exhibit diverse locomotion patterns known as gaits, which are capable of robustly traversing terrains of variable grade, roughness, and compliance. Despite the success of legs in nature, wheeled solutions still dominate the field of robotics. State-of-the-art humanoid robots have not yet demonstrated locomotion behaviors that are as robust or varied as their biological counterparts. While humans use a wide range of dynamic motions to ensure stable locomotion, most bipedal robots either reject external disturbances without changing gait or fall. Modern model-based control schemes often target individual gaits rather than realizing multiple behaviors within a single framework. Recently, researchers have proposed using the spring mass model as a compliant locomotion paradigm to create unified controllers for walking and running on bipedal systems. Although initial studies have revealed policies capable of transitioning through different gaits, most of these control laws are designed empirically using heuristics. Furthermore, these controllers have not yet been demonstrated on physical hardware, and thus their utility for real-world machines remains unclear.

This thesis investigates the optimal control of simplified bipedal point-mass models for designing unified walking and running control policies. We examine the theoretically maximum performance these models can achieve and evaluate their utility for controlling higher-order robot hardware. We hypothesize that it is not necessary to hand design these policies using heurstics. Instead, existing numerical optimization tools can generate approximate globally optimal policies, which can be used to find parametric control laws. We then attempt to transfer these low-dimensional plans onto a physical bipedal robot using a model-based controller to embed the underlying simplified model. We show that this control methodology leads to stable locomotion across several different gaits on the ATRIAS biped robot.

More Information

Thesis Committee Members:
Hartmut Geyer, Chair
Christopher Atkeson
Stelian Coros
Jan Peters, Technische Universitaet Darmstadt