Welcome to REACH

Introduction

Stroke survivors frequently experience upper-limb motor deficits that limit independence in daily tasks such as brushing teeth or reaching for objects. REACH aims to accelerate assistive technology by developing a modular RL framework that learns task policies in simulation and transfers them to a wearable robotic arm.

Our goals this year are to: model the arm and tasks in MuJoCo, train policies using algorithms such as PPO/SAC on NAU’s Monsoon cluster, and design a reusable layer of abstraction so future task and hardware revisions can be more easily integrated.

At a Glance

• Domain: Assistive Robotics, Reinforcement Learning, Simulation-to-Real
• Simulation: MuJoCo (robot + task environments)
• Training: Monsoon HPC (NAU) with PPO/SAC baselines
• Focus: Enabling stroke survivors to perform everyday tasks through adaptive reinforcement learning