1 paper
Yu Yang, Xuemeng Yang, Licheng Wen +17
Continually improving agents require dynamic interaction feedback beyond static supervision, yet direct real-environment interaction is costly, slow, unsafe, and hard to paralleliz…