Monitoring and Testing of Uncertainty and Safety for Smart Co-bots

Safe Embodied AI: 1. How can we derive meaningful prediction uncertainty bounds for AI-controlled robots in practical deployment cases in a user-friendly way, i.e., such that a non-expert end- user can specify any required
hyperparameters? 2. How can we communicate the practical consequences of said prediction uncertainty on the task outcome to an end-user, who likely has little knowledge of AI and robotics? 3. How can we build intuitive interfaces
for active learning, such that the AI can request small amounts of data from end-users in situations when it is uncertain, instead of relying on large amounts of general data? 4. How can we build robust safety guarantees into the robot controller in a way that integrates with AI decision-making algorithms and takes advantage of their ability to generalize, while still enforcing safety and task
performance?