Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore a 15-minute conference presentation introducing RoboScribe, a groundbreaking abstraction refinement-guided program synthesis framework that revolutionizes robot learning from demonstrations. Discover how this innovative approach automatically derives robot state and action abstractions from raw, unsegmented task demonstrations in high-dimensional, continuous spaces, eliminating the need for predefined domain-specific languages or user-defined state abstraction predicates. Learn about the iterative refinement process that starts with coarse abstractions and progressively enriches them until generating task-solving programs over abstracted robot environments. Understand how RoboScribe synthesizes interpretable iterative programs by inferring recurring subroutines directly from continuous state and action spaces, addressing the significant challenge of ensuring trustworthiness in neural network-based robotic systems. Examine experimental results demonstrating superior performance in terms of interpretability and efficiency, with programs that inductively generalize to long-horizon robot tasks involving arbitrary numbers of objects. Gain insights into how this framework advances programmatic reinforcement learning by making robot-control program synthesis practical in unknown environments without predefined components, presented by researchers from Rutgers University at the OOPSLA 2025 conference.
