Tail-Likelihood Reinforcement Learning

Qingyang Wu⁴, Zhongzhu Zhou⁴, Chenfeng Xu⁴

Haiwen Feng^2,3, Yuda Song¹, Aarti Singh¹, Ruslan Salakhutdinov¹

J. Andrew Bagnell^5,1, Jeff Schneider^1,†, Andrea Zanette^1,†

¹Carnegie Mellon University ²University of California, Berkeley ³Impossible, Inc. ⁴Together AI ⁵Aurora Innovation
^†Joint advising

{shrinivr, jeff4, azanette}@andrew.cmu.edu · Preprint, 2026