Optimization of Robot Configurations for Assistive Tasks

摘要

Robots can provide assistance with activities of daily
\nliving (ADLs) to humans with motor impairments. Specialized
\nrobots, such as desktop robotic feeding systems, have been
\nsuccessful for specific assistive tasks when placed in fixed and
\ndesignated positions with respect to the user. General-purpose
\nmobile manipulators could act as a more versatile form of
\nassistive technology, able to perform many tasks, but selecting a
\nconfiguration for the robots from which to perform a task can be
\nchallenging due to the high number of degrees of freedom of the
\nrobots and the complexity of the tasks. As with the specialized,
\nfixed robots, once in a good configuration, another system or the
\nuser can provide the fine control to perform the details of the task.
\nIn this short paper, we present Task-centric Optimization of robot
\nConfigurations (TOC), a method for selecting configurations for
\na PR2 and a robotic bed to allow the PR2 to provide effective
\nassistance with ADLs. TOC builds upon previous work, Task-centric
\ninitial Configuration Selection (TCS), addressing some
\nof the limitations of TCS. Notable alterations are selecting
\nconfigurations from the continuous configuration space using
\na Covariance Matrix Adaptation Evolution Strategy (CMA-ES)
\noptimization, introducing a joint-limit-weighted manipulability
\nterm, and changing the framework to move all optimization
\noffline and using function approximation at run-time. To evaluate
\nTOC, we created models of 13 activities of daily living (ADLs) and
\ncompared TOC’s and TCS’s performance with these 13 assistive
\ntasks in a computer simulation of a PR2, a robotic bed, and a
\nmodel of a human body. TOC performed as well or better than
\nTCS in most of our tests against state estimation error. We also
\nimplemented TOC on a real PR2 and a real robotic bed and
\nfound that from the TOC-selected configuration the PR2 could
\nreach all task-relevant goals on a mannequin on the bed.