# When (and How) to Trust the Expert: Diagnosing Query-Time Expert-Guided Reinforcement Learning

arXiv: https://arxiv.org/abs/2605.09109 · HTML: https://mlsanity.com/paper/2605.09109/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Deep RL algorithms and representations, one of 15 directions in Deep reinforcement learning methods.
- Builds on 23 of its 36 references that are in the map.
- First posted on arXiv 9 May 2026.

## Placed in

- Deep reinforcement learning methods / [Deep RL algorithms and representations](https://mlsanity.com/deeprl/deep-rl-algorithms-and-representations/)

## Builds on (most cited first, 23 of 23)

- [Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor](https://mlsanity.com/paper/1801.01290/) (2018)
- [Optuna: A Next-generation Hyperparameter Optimization Framework](https://mlsanity.com/paper/1907.10902/) (2019)
- [D4RL: Datasets for Deep Data-Driven Reinforcement Learning](https://mlsanity.com/paper/2004.07219/) (2020)
- [Offline Reinforcement Learning with Implicit Q-Learning](https://mlsanity.com/paper/2110.06169/) (2021)
- [DeepMind Control Suite](https://mlsanity.com/paper/1801.00690/) (2018)
- [Deep Reinforcement Learning at the Edge of the Statistical Precipice](https://mlsanity.com/paper/2108.13264/) (2021)
- [What Matters in Learning from Offline Human Demonstrations for Robot Manipulation](https://mlsanity.com/paper/2108.03298/) (2021)
- [AWAC: Accelerating Online Reinforcement Learning with Offline Datasets](https://mlsanity.com/paper/2006.09359/) (2020)
- [Deep Exploration via Bootstrapped DQN](https://mlsanity.com/paper/1602.04621/) (2016)
- [A Tutorial on Thompson Sampling](https://mlsanity.com/paper/1707.02038/) (2017)
- [Efficient Online Reinforcement Learning with Offline Data](https://mlsanity.com/paper/2302.02948/) (2023)
- [Uncertainty-Based Offline Reinforcement Learning with Diversified Q-Ensemble](https://mlsanity.com/paper/2110.01548/) (2021)
- [Residual Reinforcement Learning for Robot Control](https://mlsanity.com/paper/1812.03201/) (2018)
- [Learning to Optimize Via Posterior Sampling](https://mlsanity.com/paper/1301.2609/) (2013)
- [SUNRISE: A Simple Unified Framework for Ensemble Learning in Deep Reinforcement Learning](https://mlsanity.com/paper/2007.04938/) (2020)
- [Policy Expansion for Bridging Offline-to-Online Reinforcement Learning](https://mlsanity.com/paper/2302.00935/) (2023)
- [Why So Pessimistic? Estimating Uncertainties for Offline RL through Ensembles, and Why Their Independence Matters](https://mlsanity.com/paper/2205.13703/) (2022)
- [Imitation Bootstrapped Reinforcement Learning](https://mlsanity.com/paper/2311.02198/) (2023)
- [Conservative Bandits](https://mlsanity.com/paper/1602.04282/) (2016)
- [Conservative Contextual Linear Bandits](https://mlsanity.com/paper/1611.06426/) (2016)
- [Reinforcement Learning with Sparse Rewards using Guidance from Offline Demonstration](https://mlsanity.com/paper/2202.04628/) (2022)
- [PC-Gym: Benchmark Environments For Process Control Problems](https://mlsanity.com/paper/2410.22093/) (2024)
- [Solving Robotics Tasks with Prior Demonstration via Exploration-Efficient Deep Reinforcement Learning](https://mlsanity.com/paper/2509.04069/) (2025)

## Abstract (arXiv)

Many continuous-control problems ship with a competent but suboptimal controller (a tuned PID, a hand-designed gait). A growing family of methods uses such controllers as queryable experts during RL, but each method has been proposed in isolation, on a different benchmark, without imperfect-expert testing. We harmonize the comparison on a shared SAC backbone, common HPO and evaluation protocols, 100/50 seeds per (env, method), and a degradation sweep over expert undertuning, action bias, and observation noise. The comparison surfaces three failure modes single-paper evaluations miss: (F1) a critic blind spot under argmax-plus-bootstrap that drags IBRL below no-expert SAC on experts close to the no-expert-RL ceiling (RL-near-ceiling, distinct from the absolute physical ceiling); (F2) residual saturation on far-from-optimal experts; and (F3) warm-start buffer poisoning that collapses training-time-handoff methods under deployment-time expert undertuning. No single method dominates: each wins on one task-structure regime and fails predictably elsewhere; on RL-near-ceiling experts (FourTank, GlassFurnace) no query-time method clears the expert within our 1M-step budget, leaving open whether this is a fundamental wall or a budget effect. We convert the spread into a testable decision rule keyed on three pre-training observables (expert quality, task termination, perturbation type). The benchmark, taxonomy, and decision rule are the primary contribution; we additionally describe EDGE, a softmax-over-ensemble-LCB design point used to demonstrate that both axes the taxonomy points to (gate form, scoring rule) are individually exploitable.