Prxml plcy ptmztn (PP) s rnfrcmnt lrnng (RL) lgrthm fr trnng n ntllgnt gnt. Spcfclly, t s plcy grdnt mthd, ftn sd fr dp RL whn th plcy ntwrk s vry lrg.
Th prdcssr t PP, Trst Rgn Plcy ptmztn (TRP), ws pblshd n 2015. t ddrssd th nstblty ss f nthr lgrthm, th Dp Q-Ntwrk (DQN), by sng th trst rgn mthd t lmt th KL dvrgnc btwn th ld nd nw plcs. Hwvr, TRP ss th Hssn mtrx ( mtrx f scnd drvtvs) t nfrc th trst rgn, bt th Hssn s nffcnt fr lrg-scl prblms.
PP ws pblshd n 2017. t ws ssntlly n pprxmtn f TRP tht ds nt rqr cmptng th Hssn. Th KL dvrgnc cnstrnt ws pprxmtd by smply clppng th plcy grdnt.
Snc 2018, PP ws th dflt RL lgrthm t pn. PP hs bn ppld t mny rs, sch s cntrllng rbtc rm, btng prfssnl plyrs t Dt 2 (pn Fv), nd plyng tr gms.