The Resilient Autonomy theme is focused on taking robots out of the prototype stage and accelerating their adoption in large scale deployments. The core research directions defined in this theme will help bring about robotic systems capable of operating in wildly varying environmental conditions, in dense traffic or crowded settings, with foresight and robustness to unexpected incidents. The basic building blocks of localization and mapping, object detection and tracking, and motion planning and control must all be hardened and made both efficient and reliable throughout a robot’s operating range and lifetime. Key research areas in the Resilient Autonomy theme include:

• Robust and practical mapping and localization.Sensing and mapping the working environment, including exploring for new data, sensor selection and optimization, sensor data integration and integrity monitoring, quality of service guarantees. Leveraging massively distributed sensor networks including consumer Internet of Things (IoT) devices associated with users of the space. Incorporating deep learning methods for front-end sensor measurement correspondence, uncertainty assessment and place recognition.
• Detecting, tracking and predicting object motions.Critical to operation of self-driving cars, robotic systems in shopping centres, unmanned aerial and aquatic vehicles, accurate detection of moving objects and their motion tracking and prediction remains elusive. We will leverage detection gains in AI and deep knowledge of inference and tracking to make strides in robust tracking and motion prediction.
• Robust robot motion planning in complex environments.Motion planning methods that span robotic domains and deal with complex 6 Degree of Freedom (DOF) environments and numerous dynamic obstacles or agents. Dealing with unpredicted disturbances (e.g., road conditions, wind and wave action). Planning in the presence of uncertainty and incorrect predictions.
• Compliant manipulation.Grasping and manipulation methods for a wide range of tasks. Compliant and other novel locomotive strategies. Dealing with unpredicted disturbances (e.g., wind and wave action). Planning in the presence of uncertainty and incorrect predictions.
• Realistic, hybrid simulation generation.Creating automated methods of incorporating robotic sensor data into high fidelity robot, sensor and environment models for simulation-based evaluation of autonomy methods. This hybrid approach to modeling allows known physical constraints to remain enforced while developing data-driven simulation tools that accurately reflect specific robots.
• Robot software engineering.Devising software tools and techniques that contribute to the reliability of robot controllers. Robots are often complex distributed software systems: the most challenging type of software to create and validate. We seek to make it easier and quicker for researchers and companies to develop adequate robot software, and to encourage network added value by making advanced robot software easy to share. Our approach is to continuously share and integrate the components developed in the Resilient Autonomy theme, using Robot Operating System (ROS) as default platform, but exploring ideas for extensions, validation tools and next-generation platforms.