Optimization-based Motion Planning for Humanoid Fall Recovery
Date
2020
Authors
Advisors
Journal Title
Journal ISSN
Volume Title
Repository Usage Stats
views
downloads
Abstract
Humanoid robots are created to look like humans, behave like humans, and ultimately reason in the same level as humans. Carrying on the hopes for emulating human beings' capabilities of performing dexterously and reliably in real world scenarios, humanoids are expected to undertake a wide variety of tasks including taking care of the elderly, doing house cleaning and disinfecting (especially important during COVID-19), and conducting operations in dangerous situations such as search and rescue during fire or earthquakes. The realization of these expectations of humanoid platforms needs continuous coordination of their upper and lower limbs in challenging environments and legged robots' inherent terrain adaptability makes them competent to provide assistance in these hazardous conditions. Though being considered as attractive candidates, humanoid robots suffer from a great risk of falling resulting from a relatively high center of mass position and a limited area of region of support. This prone-to-falling characteristic of humanoids’ bipedal walking makes them much harder to control, and falls can cause costly failures. As a result, the ability to regain balance from falling is a prerequisite before humanoids can be confidently applied to execute significant tasks. Despite the rise of relevant research on humanoid fall recovery in recent decades, humanoid's self-balancing in response to unexpected disturbances in arbitrary environment remains to be a difficult problem due to humanoid's high degrees of freedom, complicated nonlinear system dynamics, and a ``real-time" computational requirement owing to falling.
This dissertation focuses humanoid fall recovery with optimization-based motion planning approach. To advance state-of-the-art recovery strategies which mainly focus on open environment, I introduce motion planning algorithms which generalize fall recovery to both open and cluttered environments. I demonstrate two main contributions in this dissertation:1. The development and implementation of an efficient motion planner which enables humanoid to recover from falling by making hand contact with walls or other surfaces in the cluttered environment. This approach extends humanoid's balancing capability to cluttered environment with making hand contact and this ability to make use of environmental object for fall prevention improves humanoids' efficiency and reliability. 2. The proposal and development of a multi-contact motion planner which generalizes humanoid fall recovery in both open and cluttered environment. This algorithm unifies existing recovery strategies, such as inertial shaping, protective stepping, and hand contact, and automatically plans one strategy or a combination of strategies to regain robot's balance based on its disturbed state and nearby environment features. By enabling humanoid to reason how to regain balance on its own, this algorithm makes a significant contribution to the improvement of humanoid's sustainability in arbitrary environment.
Overall, these contributions advance state-of-the-art humanoid technologies with the ability to 1). use hand contact for fall prevention in cluttered environment and 2). reason how to regain balance in both open and cluttered environments. By further enhancing legged machines' capability of self-balancing, methods discussed in this dissertation have the potential to realize a more effective and more reliable humanoid performance in real world.
Type
Department
Description
Provenance
Citation
Permalink
Citation
Wang, Shihao (2020). Optimization-based Motion Planning for Humanoid Fall Recovery. Dissertation, Duke University. Retrieved from https://hdl.handle.net/10161/22157.
Collections
Except where otherwise noted, student scholarship that was shared on DukeSpace after 2009 is made available to the public under a Creative Commons Attribution / Non-commercial / No derivatives (CC-BY-NC-ND) license. All rights in student work shared on DukeSpace before 2009 remain with the author and/or their designee, whose permission may be required for reuse.