Progress Reward Modeling for Robotic Learning: A Comprehensive Survey

Researchers published a survey that frames progress reward modeling as a better alternative to sparse terminal success signals in robotic learning.

The paper organizes the field by interface design, internal modeling approaches, and the data and benchmarks used to train and test methods.

It argues that intermediate progress feedback can improve robot behavior guidance and make learning more effective.

The survey also highlights evaluation gaps, current limitations, and promising future directions for the area.