Content

Speaker

Nisal Perera

Abstract

Human locomotion relies on mechanical coordination across the leg and foot. Muscles spanning multiple joints distribute forces and transfer mechanical energy between joints, while toe articulation and foot stiffness shape how the body pushes against the ground. Human experiments demonstrate that reducing foot-muscle function diminishes push-off and increases compensatory work at the hip, while studies of ankle–foot systems show that toe stiffness changes the work delivered during push-off. These findings motivate robotic systems that incorporate coordinated actuation and articulated toes, but leave open how these features should be controlled and whether their mechanical advantages translate into lower electrical energy consumption.

This thesis investigates the realization and evaluation of these biomechanical principles in humanoid robots. Building upon the mechanisms introduced in HyperLeg, our robots are designed to leverage co-actuation — coupling actuator across multiple joints rather than pairing each actuator with a single, so that their efforts can combine during coordinated knee extension and ankle and toe push-off. This architecture allows multiple actuators to contribute to propulsion and augment the torque available at individual joints. However, while our robotic legs can push against the ground harder than legs with decoupled, open-chain actuation mechanism, the coupling makes motion control complicated, as joint behavior depends on configuration and on feedback from multiple actuators. The research therefore identifies the proper manner to achieve desired joint behavior — impedance — by regulating actuator impedance and feedforward input to enable predictable motion and ground interaction.

To address these challenges, StaccaToe and PresToe provide single-leg and humanoid platforms for experimental investigation. The research will evaluate how actuator feedback and feedforward commands can be coordinated to achieve desired joint behavior across changing leg configurations. Because this behavior also depends on transmission dynamics, actuator response, signal processing, and feedback delays, the proposed work will characterize these effects and incorporate them into simulation. Comparisons between simulated and measured hardware responses will assess model accuracy, while controller and learned-policy deployment will test whether accounting for these dynamics improves transfer from simulation to the physical robot.

Using this experimentally grounded modeling and control approach, the research will compare active-toe and constrained-toe configurations to investigate how toe use affects walking efficiency and motion capability within the co-actuated architecture. Measurements of electrical energy consumption, actuator loading, and contact-transition mechanics will identify the conditions under which toe use provides benefits and explain the factors that limit them. The anticipated outcome is a clearer understanding of how biomechanically inspired mechanisms, their control, and their physical dynamics jointly determine humanoid performance, supported by experimentally evaluated approaches for realizing these capabilities in robotic systems.

Advisor

Donghyun Kim