Robots and robotics are becoming increasingly important outside their normal industrial applications, and are seen as having enormous potential for social good, as well as industrial and economic value. The research project at the heart of this career training and development proposal will explore real-time developmental learning in a humanoid robot on practical real-world manipulation tasks. Such learning is essential if robots are to fulfil their promise for improving lives and living.The learning will be driven by the complete high-dimensional sensorial context of the robot, not just a task-specific subset of sensors and pre-established features. The principal scientific objectives of this project are to study how the full sensorial context of an embodied robot platform can be used to drive behaviour; how such behaviour can be self-organized in real-time to enable learning in dynamic (and learning) contexts; and how to not only ground existing symbols in the sensorimotor interaction, but also adapt the structure of the neural substrate to enable a development of the set of symbols used. These objectives will be achieved through theoretical studies concerning the relationship of a robot’s “state” and a robot’s sensed context; through an open source software implementation of a developmental learning that associates high-dimensional sensor input to actuation in real-time; and through controlled experimental evaluations with the iCub humanoid robot and human teachers. Findings will be openly published to allow replication, verification and refinement by the research community; stimulating commercial development and adoption of useable technology.