# Reflective Prompted Policy Optimization: Trajectory-Grounded Revision and Salience Bias

arXiv: https://arxiv.org/abs/2605.08315 · HTML: https://mlsanity.com/paper/2605.08315/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 16 of its 22 references that are in the map.
- First posted on arXiv 8 May 2026.

## Builds on (most cited first, 16 of 16)

- [Proximal Policy Optimization Algorithms](https://mlsanity.com/paper/1707.06347/) (2017)
- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [Reflexion: Language Agents with Verbal Reinforcement Learning](https://mlsanity.com/paper/2303.11366/) (2023)
- [Self-Refine: Iterative Refinement with Self-Feedback](https://mlsanity.com/paper/2303.17651/) (2023)
- [Continuous control with deep reinforcement learning](https://mlsanity.com/paper/1509.02971/) (2015)
- [Trust Region Policy Optimization](https://mlsanity.com/paper/1502.05477/) (2015)
- [Addressing Function Approximation Error in Actor-Critic Methods](https://mlsanity.com/paper/1802.09477/) (2018)
- [Asynchronous Methods for Deep Reinforcement Learning](https://mlsanity.com/paper/1602.01783/) (2016)
- [Large Language Models as Optimizers](https://mlsanity.com/paper/2309.03409/) (2023)
- [Reward Design with Language Models](https://mlsanity.com/paper/2303.00001/) (2023)
- [DrEureka: Language Model Guided Sim-To-Real Transfer](https://mlsanity.com/paper/2406.01967/) (2024)
- [TreeRL: LLM Reinforcement Learning with On-Policy Tree Search](https://mlsanity.com/paper/2506.11902/) (2025)
- [Large Language Model as a Policy Teacher for Training Reinforcement Learning Agents](https://mlsanity.com/paper/2311.13373/) (2023)
- [TSR: Trajectory-Search Rollouts for Multi-Turn RL of LLM Agents](https://mlsanity.com/paper/2602.11767/) (2026)
- [Enhancing Q-Learning with Large Language Model Heuristics](https://mlsanity.com/paper/2405.03341/) (2024)
- [Prompted Policy Search: Reinforcement Learning through Linguistic and Numerical Reasoning in LLMs](https://mlsanity.com/paper/2511.21928/) (2025)

## Abstract (arXiv)

Existing LLM-based policy optimizers see only scalar rewards: that a policy scored 0.45, but not whether the agent got stuck in a loop, fell into a hole on the third step, or performed well on 19 out of 20 rollouts and failed catastrophically on one. We propose Reflective Prompted Policy Optimization (R2PO), a two-stage LLM framework for policy search over compact policy classes that augments scalar reward feedback with trajectory-level behavioral evidence. A Search-LLM proposes candidate policy parameters; the environment executes them; a Critic-LLM inspects the resulting rollouts and proposes targeted revisions grounded in observed states, actions, and rewards. Across ten environments, ablations show R2PO's gains require separating global search from behavior-grounded revision and using selection to filter high-variance edits. We further identify a dominant failure mode, salience bias: when presented with multiple rollouts, the Critic-LLM fixates on improving a single failure even when most trajectories succeed. In a three-trajectory variant where the Critic-LLM sees the best, worst, and median rollout, this behavior explains 76.6% of regressions on CartPole. R2PO mitigates this by reasoning over aggregate rollout statistics, median-trajectory selection, and a revision rule. Using a 20B open-weight model, R2PO achieves the highest mean best reward across all ten environments, reaches near-optimal performance substantially earlier (e.g., near-maximum CartPole reward within ~500 episodes), and trains far more stably than both deep RL and prior LLM-based methods. These results show that treating trajectories as first-class in-context evidence, rather than artifacts reduced to scalar returns, changes how even comparatively small LLMs search over policy spaces, enabling them to learn faster, diagnose more precisely, and reliably improve external controllers.