Exo-Plore: Exploring Exoskeleton Control Space through Human-aligned Simulation
Geonho Leem, Jaedong Lee, Jehee Lee, Seungmoon Song, Jungdam Won
TL;DR
Exo-plore is a simulation framework that uses neuromechanical modeling and deep reinforcement learning to optimize exoskeleton assistance, reducing the need for extensive human trials and accommodating pathological gaits.
Contribution
It introduces a novel simulation-based approach for exoskeleton control optimization that captures human adaptation and generalizes to pathological conditions.
Findings
Generates realistic gait data reflecting human adaptation.
Produces reliable optimization results despite gait variability.
Generalizes to pathological gaits with strong correlations.
Abstract
Exoskeletons show great promise for enhancing mobility, but providing appropriate assistance remains challenging due to the complexity of human adaptation to external forces. Current state-of-the-art approaches for optimizing exoskeleton controllers require extensive human experiments in which participants must walk for hours, creating a paradox: those who could benefit most from exoskeleton assistance, such as individuals with mobility impairments, are rarely able to participate in such demanding procedures. We present Exo-plore, a simulation framework that combines neuromechanical simulation with deep reinforcement learning to optimize hip exoskeleton assistance without requiring real human experiments. Exo-plore can (1) generate realistic gait data that captures human adaptation to assistive forces, (2) produce reliable optimization results despite the stochastic nature of human…
Peer Reviews
Decision·ICLR 2026 Poster
- Clear problem framing and a convincing argument for simulation‑first controller selection when HILO is impractical. (Intro; Fig. 1.) - Well engineered pipeline (PoseNet/MCN/IMR; LHS + Huber + gradient penalty) with thoughtful ablations and appendices documenting design choices. (Method; Appendix B, D-G). - Clinically relevant analysis of pathological gaits
- The paper’s scope seems more suitable to a robotics or rehabilitation venue than here. While applications to a particular domain, such as, rehabilitation and robotics are a perfect fit for this venue, the core of this work lies primarily in the exoskeleton control and simulation framework rather than in a methodological advance in/applying deep learning itself. - Fitting (α,β) to match the human CoT-speed parabola incorporates experimental behavior into the simulator (Algorithm 1–2). This aid
The paper is well written and organized. The problem is well motivated by the authors who position it well among existing literature. Bridging the gap between neuromuscular simulation and reality is a highly relevant open problem in robotics. The paper also presents interesting formulations - The novel reward term r_{hei} for training the generative GaitNet, intra-muscular regularizer in the loss term to learn Muscle Coordination Network, and the Surrogate Network which estimates cost of trans
As the authors themselves acknowledge, while the results here show promise in closing the sim-2-real gap in neuro-mechanical simulation, the ultimate test is how well this actually works in real world testing, which is still missing. So, a question still remains as to how effective the proposed methodology is? But this does not take away the good work and its contributions in this paper. Small observations - In equation 16, it's not clear what M and n stand here. There are a lot of variables d
1. The Exo-plore simulation-based framework represents a notable step forward in the development of biomechanical simulation environments that reduce the sim2real gap. The framework demonstrated promising results when benchmarked against experimental data despite the complexity of the musculoskeletal system. 2. The possibility to simulate assistive devices in combination with biomechanical models is also another promising result. Fine-tuning exoskeletal control parameters is often hard, and Exo
1. Authors could improve the literature on human-in-the-loop optimization for exoskeletons. The group of Prof. Elliott Rouse at the U. Michigan has done extensive work on the topic on both prostheses and exoskeletons, but I could only find 1 work from his group in your references. Consider extending your literature review and include also this relevant portion of literature. 2. In neuromechanical simulations, authors should briefly mention the fact that Hill-type muscle actuators are an approxi
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsProsthetics and Rehabilitation Robotics · Stroke Rehabilitation and Recovery · Balance, Gait, and Falls Prevention