Date of Award

6-2026

Document Type

Thesis

Publisher

Santa Clara : Santa Clara University, 2026

Degree Name

Master of Science (MS)

Department

Electrical and Computer Engineering

First Advisor

Maria Kyrarini

Abstract

Robots are now able to replicate the physical performance or even surpass that of a human, but are still limited in their ability to operate in unstructured environments. For these unstructured applications, human teleoperators can prove useful for controlling multirobot systems in highly dynamic and potentially dangerous environments such as chemical and industrial facilities. Teleoperation of complicated robots with several degrees of freedom, like robotic manipulators, traditionally requires immense skill and coordination to ensure safe and efficient task completion. Intent inference offers a promising path to reduce human operator burden by enabling the robotic system to anticipate and assist with operator goals.

This work builds on GUIDER, Global User Intent Dual-phase Estimation for Robots, which is an example of a goal-free probabilistic framework for human intent inference during robotic manipulation. Intent inference frameworks are used to predict the operator’s goals and intentions based on their inputs and the current state of the environment, allowing the robot to assist in a more proactive manner. Goal-free intent inference does not require a predefined set of goals or tasks. This approach is particularly useful in unstructured environments where the operator may need to perform a wide range of tasks that cannot be easily categorized.

Through this work, we made four distinct contributions: a custom remote teleoperation interface for the Franka Emika Panda, four adaptations to make the GUIDER manipulation phase functional on physical robot data including a grasping mode that prioritizes feasible grasp regions, an evaluation of GUIDER against a dataset of teleoperation data that we collected from the Franka Emika Panda robot arm across three manipulation scenarios, and all associated documentation of the engineering and design decisions made in the development of the teleoperation interface and the adaptations to the GUIDER framework, providing insights and guidance for future work.

To support operation, we added online probability updates, workspace limits, support-plane filtering, and a grasping mode that prioritizes feasible grasp regions, all of which are tested on the recorded data while preserving its original temporal conditions. Across 20 manipulation steps in three scenarios, GUIDER kept the human intent within the correct grasp-candidate set in all cases, achieving a time to confident prediction of 3.7 s and a prediction stability of 96.4%. This shows the potential of GUIDER to enhance teleoperation by providing timely and accurate intent predictions, improving the efficiency and safety of robotic manipulation in unstructured environments. Collecting real-world data allowed us to identify areas for improvement in the framework, like more robust handling of noisy inputs and dynamic environments which are common in real-world applications.

Available for download on Wednesday, September 01, 2027

Share

COinS