Preference-Based Learning for Exoskeleton Gait Optimization

Abstract

This paper presents a personalized gait optimization framework for lower-body exoskeletons. Rather than optimizing numerical objectives such as the mechanical cost of transport, our approach directly learns from user preferences, e.g., for comfort. Building upon work in preferencebased interactive learning, we present the COSPAR algorithm. COSPAR prompts the user to give pairwise preferences between trials and suggest improvements; as exoskeleton walking is a non-intuitive behavior, users can provide preferences more easily and reliably than numerical feedback. We show that COSPAR performs competitively in simulation and demonstrate a prototype implementation of COSPAR on a lower-body exoskeleton to optimize human walking trajectory features. In the experiments, COSPAR consistently found user-preferred parameters of the exoskeleton's walking gait, which suggests that it is a promising starting point for adapting and personalizing exoskeletons (or other assistive devices) to individual users.

Human Preference-Based Learning for High-dimensional Optimization of Exoskeleton Walking Gaits

“…The LINECOSPAR algorithm (Alg. 1) builds upon COSPAR [4] by learning a Bayesian model over users' preferences in higher-dimensional spaces. Drawing inspiration from the LINEBO algorithm [13], LINECOSPAR exploits low-dimensional structure in the search space by dividing the problem into a series of one-dimensional subproblems.…”

“…Human-in-the-loop online learning techniques have demonstrated significant potential in human-robot interaction tasks [1]- [3], such as in improving the performance of robotic assistive devices. In particular, online learning from human feedback can help to optimize walking gaits for lower-body exoskeletons [4]- [6], which are placed over existing limbs to assist mobility-impaired individuals.…”

“…We rely on users' relative preferences to learn exoskeleton gaits optimizing their comfort, as several studies have shown that for subjective human feedback, pairwise preferences are more reliable than numerical scores [10]- [12]. While interactive preference-based learning methods have been applied to robotics [4], [6], existing online preference-based learning methods are restricted to low-dimensional domains due to computational limitations; for example, previous work on preference-based exoskeleton gait optimization either learns over at most two dimensions [4] or utilizes domain knowledge to narrow the search space before performing online learning [6].…”

“…We present LINECOSPAR, a high-dimensional human preference-based learning approach that integrates existing techniques for preference learning [4] and high-dimensional optimization [13] into a unified framework. LINECOSPAR relies on preference feedback to iteratively explore onedimensional subspaces.…”

A Smart Robotic Walker With Intelligent Close-Proximity Interaction Capabilities for Elderly Mobility Safety

“…We argue that intelligence is essential for an elderly walker to detect abnormal user behaviors and provide timely safety support, since primitive assistance devices, such as rollators and walkers, are much likely to fail (Bertrand et al, 2017). Exoskeleton (Tucker et al, 2019) is another approach with multiple robotic joints and links worn onto the user body, effective but less practical for daily wearing by older persons. Moreover, rather than merely using remote button (Glover, 2003), voice (Gharieb, 2006), or gesture (Gleeson et al, 2013) to achieve user interaction, older persons need various modes of human-robot interaction for convenience and efficiency.…”