Preprint
Oct 2026
One Policy, Many Styles: Where to Aggregate Preferences for Steerable Humanoid Control?
This evaluation combines Preference Reward Responsiveness (PRR) with coverage and utility metrics to compare reward-level, advantage-level, loss-level, and PCGrad-based aggregation across two humanoid-control frameworks.
Zi-Qian Huang, Yu-Zhuo Chen, Shi-Min Zhang et al.
· 0 citations