LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation
arXiv cs.AI · 16 Oct 2026, 03:59 AEST
Personalization is essential for general-purpose robots to transition into household environments. The LIBERO-PeRM benchmark aims to enhance personalized robotic manipulation by studying various aspects of preference learning in embodied agents.
Key facts
- While general robotic manipulation has been extensively studied, personalized robotic manipulation (PeRM) lacks a controlled benchmark for separating task completion from preference satisfaction.
- LIBERO-PeRM studies five questions: 1) how policies follow preferences in zero-shot settings; 2) how they learn single preferences from demonstrations; 3) how efficiently they acquire preferences as demonstrations increase; 4) how learned preferences generalize across layouts and tasks; and 5) how policies compose preferences and behave under conflicts.
- Experiments show that current VLA policies struggle with zero-shot personalization, while SFT makes preferences learnable.
Read aloud
Personalization is essential for general-purpose robots to transition into household environments. The LIBERO-PeRM benchmark aims to enhance personalized robotic manipulation by studying various aspects of preference learning in embodied agents. While general robotic manipulation has been extensively studied, personalized robotic manipulation PeRM lacks a controlled benchmark for separating task completion from preference satisfaction. LIBERO-PeRM studies five questions: one how policies follow preferences in zero-shot settings; two how they learn single preferences from demonstrations; three how efficiently they acquire preferences as demonstrations increase; four how learned preferences generalize across layouts and tasks; and five how policies compose preferences and behave under conflicts. Experiments show that current VLA policies struggle with zero-shot personalization, while SFT makes preferences learnable.
Sources in this cluster
- LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation
arXiv cs.AI · 16 Oct 2026, 03:59 AEST