What is developmental robotics and how does it differ from other AI fields?
Developmental robotics, also called epigenetic robotics, is a scientific field that studies how embodied machines can learn new skills and knowledge continuously across a lifetime, starting from little or no prior specification of what to learn. It differs from classical artificial intelligence by focusing on embodied sensorimotor and social skills rather than abstract symbolic reasoning, and from cognitive robotics by targeting the processes that form cognitive capabilities rather than the capabilities themselves.
Who first proposed the ideas behind developmental robotics?
Alan Turing and other pioneers of cybernetics formulated the core questions and general approach as early as 1950. Systematic scientific investigation of those questions did not begin until the end of the twentieth century.
What skills do developmental robots try to learn?
Developmental robots are designed to acquire three broad categories of skills mirroring infant development: sensorimotor skills such as hand-eye coordination, locomotion, and tool use; social and linguistic skills including turn-taking, lexicons, and grammar grounded in physical experience; and cognitive skills such as self/non-self distinction, attention, categorization, and rudimentary theories of mind.
What is curiosity-driven learning in developmental robotics?
Curiosity-driven learning refers to a class of intrinsic motivations that push a robot to seek novelty, challenge, or learning progress for its own sake, rather than for any externally specified reward. It is one of several motivational mechanisms studied in the field to guide a robot's exploration without programming in every task in advance.
What was the first major international conference on developmental robotics?
The first international meeting devoted to the computational understanding of mental development by robots and animals was the NSF and DARPA funded Workshop on Development and Learning, held on the 5th-the 7th of April 2000 at Michigan State University. Two subsequent separate conferences later merged into the ICDL-EpiRob conference in 2011.
What are the biggest unsolved challenges in developmental robotics?
The field identifies several open challenges: no experiments have yet run longer than a few days despite human infants needing years to acquire basic skills; high-dimensional sensorimotor spaces remain a technical obstacle; and the interaction of multiple learning mechanisms such as intrinsic motivation, social guidance, and maturational constraints has not been studied systematically. The symbol grounding problem, how compositionality and hierarchical structure emerge during development, is also unresolved.