Interface-Aware Teleoperation Assistance
Overview
Developed algorithms for providing user-customized, task-agnostic assistance during high-DoF robot teleoperation (Kinova Jaco robotic arm) using a low-dimensional control interface (Sip/Puff device), based on user intent estimation using LSTMs and GRUs.
Main contributions:
- An end-to-end pipeline (data collection → model training → rollout of assistance), implemented on hardware and evaluated in a preliminary human subjects case study with 3 SCI participants.
- A reformulated algorithm and implementation to enhance practicality for task- and environment-agnostic applications, allowing use in changing environments — evaluated in a human subjects experiment with 8 SCI participants.
- Deployment of the teleoperation framework as an assisted control and training interface, enabling operators to collect high-quality demonstration data for robot manipulation and activities-of-daily-living tasks while receiving real-time control assistance.
