SPEAKER: Prof. Nico Hochgeschwender, University of Bremen
ABSTRACT: Defining acceptable robot behavior is challenging. A vague requirement like “the robot shall pick and place the object safely and reliably” is not directly executable; it depends on task context, object properties, robot embodiment, environmental conditions, timing, tolerances, and observations during execution. This makes acceptance testing a scientific challenge in robotics, requiring novel methods to bridge high-level validation intent and repeatable, executable tests. In this talk, I will present our recent work on RobBDD, a domain-specific language for expressing robotic acceptance criteria in a behavior-driven style. The language allows us to express tasks, initial conditions, scenario variations, and expected outcomes in a human-readable form structured enough for transformation into executable tests. A central part of the approach is a knowledge graph representation that explicitly captures acceptance criteria, scenario elements, and their relationships, making the specifications queryable, reusable, and traceable. Using a pick-and-place application as an example, I show how these specifications connect to our tooling for generating and executing simulation-based test cases. The talk will illustrate how acceptance testing can make robot validation more explicit and systematic: instead of asking whether a robot succeeds in a selected setup, we can ask under which variations of robots, objects, poses, and environments the intended behavior holds, where it fails, and how such failures relate to concrete acceptance criteria.