Back to AI News
arxiv

Sampling-Based Control Can Exploit Human-Preference Rewards, Cutting True Return by up to 76%

This paper investigates how reward models learned from human preferences can be exploited by sampling-based control methods like MPPI and best-of-N selection. Experiments on HalfCheetah, Hopper, and Meta-World tasks show that harder search can reduce true return by up to 76%, with drawer-open failing in all 20 runs despite high held-out label accuracy. The authors propose a chi-squared-regularized MPPI rule calibrated from oracle rankings that preserves 86-94% of cloned policy return, though it is cautious where search helps.

AI Brief

Provides concrete evidence and a practical fix for a critical failure mode in preference-based reward optimization, relevant to robotics and RL safety.

arXiv:2611.09670v1 Announce Type: cross Abstract: Robots increasingly choose actions by searching against a reward model learned from human preferences, in model-predictive control or best-of-$N$ selection; we measure what this does to the true objective. On HalfCheetah and Hopper with crowd-sourced labels, and on Meta-World drawer-open and plate-slide with synthetic labels and blind labels from the two authors, harder MPPI search cuts true return by up to 76%: with human labels the drawer opens in 0 of 20 runs against 10 of 10 for the cloned policy the planner starts from, while held-out label accuracy stays at 0.75-0.97. On plate-slide human labels are exploited less than synthetic ones, and on window-open search helps. True return roughly follows how far the planner's weights move from its proposals ($\sqrt{\mathrm{KL}}$) for both weighting rules we test, and best-of-$N$ with a verifier from synthetic labels falls below a single noisy action. We prove that $\chi^2$-regularized MPPI at a temperature set by the scorer's error spread on the proposals is worst-case optimal, that no positive gain can be certified once that spread exceeds the scorer's own, and that an episode's return splits exactly into per-step terms, of which the guarantee covers the selection terms. Calibrated from 100 proposal pairs ranked by true return (an oracle) on start states the evaluation never uses, this rule is the safest fix we test: with each of five independent sets it succeeds in every arm run and keeps on average 86-94% of the cloned policy's return on plate-slide and drawer-open, though it is cautious where search helps. Calibration from a person's comparisons cannot detect a bias the scorer learned from that person; correcting for the pairs people cannot tell apart, or a second person, limits the damage.
Be the first to rate this article.

Comments & Questions (0)

Captcha
Please be respectful — let's keep the conversation friendly.

No comments yet

Be the first to comment!