PUB - Publikationen an der Universität Bielefeld

User assistants aim to support users in complex tasks. An assistant has to handle varying user skills, different environments, as well as occurring deviations. Therefore, it is crucial that an assistant adapts to the current situation and the environment in order to provide customized assistance. A highly dynamic field of great relevance for the user is the cooking domain, as cooking is an Instrumental Activity of Daily Living. When a user prepares a menu, the combination of multiple recipes and the synchronization of servings time is crucial for success. With different available devices in private kitchens as well as variances in cooking skills of the users, the concurrent consideration of multiple adaptation layers is required in such a dynamic process.
In this thesis, the HanKA cooking assistant was developed. The assistant includes solutions that allow the free combination and synchronization of recipes, the consideration and control of the available devices, the adaptation to the cook, and the ongoing process adaptation. While these adaptations are implemented and evaluated in the cooking domain, they are transferable to user assistance in general. In order to provide adaptive assistance, a decision base is required. Therefore, a distributed knowledge representation is proposed that enriches recipes with additional knowledge from ingredient representations and suggested action_templates. Since the devices available in a kitchen vary, not all recipes can be prepared in each kitchen. In order to enable environmental awareness, the available devices with their abstracted functionalities are detected, controlled, and utilized by the assistance. Here, the abstracted functionality of the devices is mapped to the knowledge representation. Thus, devices are controllable by the assistance in regard to the recipe. By the proposed abstract interaction types, the complete user-device interaction is supported and tracked. This enables an automatic detection of the recipe progress. In order to provide assistance adapted to the cooking skills of the user, the estimated task duration and the level of detail for the instructions are aligned with the generalized user skills. These cooking skills are stored in regard to the recipe, ingredient-action pair, and action. This enables that the assistant can consider the user experience and estimate the task duration for recipes the cook has never cooked before. Since a meal usually consists of multiple components, the assistant enables the free combination and synchronization of multiple recipes. By utilizing the enriched recipes which include required devices and user workload for each task as well as logical connections between tasks, the recipe combination is considered as an optimization problem. Here, the available devices and registered users are considered. While executing the scheduled recipes, deviations are detected and solved in order to consistently provide a plan adapted to the current situation.
In order to evaluate the proposed concurrent adaptations, the HanKA cooking assistant was successfully evaluated in a study with 12 participants. Afterwards, the proposed concepts were transferred to user assistance in general. Although each adaptation layer, environment, user, process combination, and progress detection provides value on its own, the combination and seamless integration into the assistant unfolds their full potential.