This doctoral thesis addresses the problem of increasing autonomy and robustness in robotic assembly under conditions typical for small and medium-sized enterprises (SMEs), where high variability and limited engineering resources hinder traditional automation. The work focuses on force/torque-based perception and control as a primary sensing modality for contact-rich manipulation tasks.
Three complementary research problems are investigated. First, a novel framework for snap-fit assembly outcome classification is proposed. Instead of binary success detection, the method introduces a structured taxonomy of assembly outcomes and develops a machine-learningbased classification pipeline grounded in force/torque signal analysis. Second, a new method for contact localization using wrist-mounted force/torque sensing is developed. The approach combines residual wrench compensation, geometric inference, and probabilistic state estimation to localize contact on manipulated objects under realistic sensing constraints. Third, a force-guided navigation algorithm is introduced, enabling robots to progress toward task goals while constrained by environmental contact, integrating contact localization, incremental contact mapping, and reactive motion planning.
All proposed methods are validated experimentally on real robotic platforms. The experiments demonstrate that the approaches operate consistently under the deployment constraints and provide a foundation for further development of contact-aware robotic systems in industrial applications.