![Thumbnail](/bitstream/handle/20.500.11850/705078/doctoral_thesis_langerak.pdf.jpg?sequence=10&isAllowed=y)
Open access
Author
Date
2024Type
- Doctoral Thesis
ETH Bibliography
yes
Altmetrics
Abstract
The balance between user agency and system automation in interactive intelligent systems is crucial for intuitive and efficient interactions. While fully automated systems could potentially offer greater efficiency and demonstrably improved performance, making them perfect is notoriously hard. The inevitable shortcomings of automated systems diminish usability and overall experience, thereby compromising users' perceived self-determination. Conversely, tools, systems that rely entirely on user agency and have no level of automation, though offering full control to the user, can be inefficient and fail to enhance the user's capabilities. Hence, for effective human-AI interactions, we need to find a balance between user agency and system automation. The question we address in this dissertation is "How can we balance user agency and system automation for the interaction with intelligent systems?"
We approach this challenge through four main contributions. First, we introduce a novel spherical electromagnet capable of generating adjustable forces on an untethered tool, allowing users to feel grounded forces while maintaining full agency. Second, we develop an integrated sensing and actuation system that tracks a passive magnetic tool in 3D space while simultaneously delivering haptic feedback, eliminating the need for external tracking. Third, we propose an optimal control method for electromagnetic haptic guidance systems that balances user input and system control, allowing users to adjust trajectories and speed as needed. Finally, we present a model-free reinforcement learning approach for adaptive user interfaces that learns interface adaptations without relying on heuristics or real user data.
Our findings, based on simulations and user studies, suggest that the shared control of intelligent systems has the potential to significantly outperform naive control strategies. Thus, we contribute methodologies that find an agency-automation trade-off and pave the way for more interaction with intelligent systems. Our research demonstrates that integrating models of human behavior, either explicitly or implicitly, into control strategies enables intelligent systems to better account for user agency. We show that the trade-off between user agency and system automation is not solely an algorithmic problem but must also be considered in the engineering of physical devices and interface design. We advocate for an integrated end-to-end approach to interaction with intelligent systems that incorporates algorithmic, engineering, and design perspectives. Show more
Permanent link
https://doi.org/10.3929/ethz-b-000705078Publication status
publishedExternal links
Search print copy at ETH Library
Publisher
ETH ZurichSubject
Human-Computer Interaction (HCI); Human-AI collaboration; Optimal Control and Optimization; Reinforcement Learning; HapticsOrganisational unit
09649 - Holz, Christian / Holz, Christian
More
Show all metadata
ETH Bibliography
yes
Altmetrics