Robots often miss user preferences without extra training
A new test checks if robots can follow user tastes as well as finish tasks, not just complete them.
- Publication
- arXiv
- Stage
- Preprint
- What we read
- Summary of the paper
- Authors
- Zhixu Li, Keqian Tang, Litian Gong, Jingyu Yao, Wenqian Zhang, Tianze Xu, Zehao Wang, Yuping Wang, Junge Zhang, Jiachen Li
- Universities and research institutions
- Not yet supplied in verified metadata; the Brief does not guess.
What they did and found
A preference-aware robot test shows current systems can finish tasks but often ignore stated user preferences. Supervised tuning helps learn preferences, but may reduce how smoothly actions are performed. More varied tasks and including no-preference tasks aid learning, yet adapting to new tasks and conflicting desires remains limited.
Why it matters
Plan for both task success and sticking to user tastes; mixed training can keep general abilities while handling conflicts with clear prompts.
The study separates task completion from how it is done; many systems miss preferences in new tasks, but example-based training improves adherence, sometimes at the cost of motion smoothness.
Using more varied task layouts and some no-preference data helps robots follow tastes without losing general ability, but new layouts or two conflicting preferences still underperform.
What remains uncertain
Findings come from simulations with language-described preferences; real-world use and non-language cues are not covered.
Original sources · 1
- LIBERO-PeRM: Benchmarking Personalized Robotic Manipulation ↗arXiv · 2026-10-15
Check the original paper for its authors, methods, version and access terms.
What is your take?
Ask a question, add useful context or share a different perspective. Keep the conversation respectful and grounded.