AI Paper Atlas

Research map / Reinforcement learning / RL from feedback

Polychromic Objectives for Reinforcement Learning

Jubayer Ibn Hamid, Ifdita Hasan Orney, Ellen Xu, Chelsea Finn, Dorsa Sadigh

ICLR 2026 · Reinforcement learning › RL from feedback · approach: RL fine-tuning of generative models

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
replace PPO with polychromic PPOPPO · BabyAI/MiniGrid Gotosuccess rate (%): 46.2 -> 80.2Improved
replace PPO with polychromic PPOPPO · BabyAI/MiniGrid Pickupsuccess rate (%): 33.4 -> 63.2Improved
replace PPO with polychromic PPOPPO · BabyAI/MiniGrid Bosslevelsuccess rate (%): 38.8 -> 45.2Improved
replace PPO with polychromic PPOPPO · BabyAI/MiniGrid Synthseqsuccess rate (%): REINFORCE 45.4 vs Poly-PPO 47.0Mixed
add UCB bonus to PPOPPO · BabyAI/MiniGridsuccess rate (%): Pickup 33.4 -> 27.8; Goto 46.2 -> 47.4Mixed
add UCB bonus to Poly-PPOPoly-PPO · BabyAI/MiniGridsuccess rate (%): Pickup 63.2 -> 65.6; Goto 80.2 -> 76.2Mixed
replace PPO with polychromic PPO under large initial-state perturbationsPPO · BabyAI/MiniGrid Goto, perturbedpass@1 (%): 21.1 -> 60.6Improved
replace PPO with polychromic PPO under large initial-state perturbationsPPO · BabyAI/MiniGrid Four Rooms, perturbedpass@1 (%): 15.3 -> 88.7Improved
use PPO instead of the pretrained policy for Four Rooms perturbed startsPPO · BabyAI/MiniGrid Four Rooms, perturbedpass@1 (%): 65.0 -> 15.3Hurt
apply PPO / REINFORCE fine-tuningPPO and REINFORCE · BabyAI pass@k at large kpass rateHurt
use polychromic PPOPoly-PPO · Algorithmic Creativity triangle discoverycreativity and diversityImproved
use polychromic PPO instead of PPOPPO · Algorithmic Creativity triangle discoveryvalidityHurt

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy