How to Peel with a Knife: Aligning Fine-Grained Manipulation with Human Preference
How to Peel with a Knife: Aligning Fine-Grained Manipulation with Human Preference
This research presents a learning framework for fine-grained manipulation tasks characterized by implicit success criteria, using knife peeling as a case study. The approach employs a two-stage pipeline: first, a robust initial policy is learned via force-aware data collection and imitation learning to generalize across object variations; second, the policy is refined through preference-based fine-tuning using a reward model that integrates quantitative metrics with qualitative human feedback. Using only 50 to 200 peeling trajectories, the system achieves 100% success on cucumbers, apples, and potatoes, with performance improvements of up to 40% after fine-tuning. Policies trained on one produce category also transfer zero-shot to held-out categories, achieving 50% success on zucchini, 90% on pear, and 80% on daikon radish.
The study introduces a two-stage framework combining force-aware imitation learning with preference-based fine-tuning to address continuous, subjective success criteria in contact-rich manipulation tasks like peeling.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Control
- Reported autonomousOpen source S2
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
arXiv:2603.03280v2 Announce Type: replace Abstract: Many essential manipulation tasks - such as food preparation, surgery, and craftsmanship - remain difficult for autonomous robots. These tasks are characterized not only by contact-rich, force-sensitive dynamics, but also by their "implicit" success criteria: unlike pick-and-place, task quality in these domains is continuous and subjective (e.g. how well a potato is
Open source S2
etuning using a learned reward model that combines quantitative task metrics with qualitative human feedback, aligning policy behavior with human-evaluated task quality. Using only 50~200 peeling trajectories, our system achieves 100% same-produce success on cucumbers, apples, and potatoes, with performance improving by up to 40% through preference-based finetuning. Remarkably, policies trained on one produce categor
Open source S4
y also transfer zero-shot to selected held-out produce categories, achieving 50% success on zucchini, 90% on pear, and 80% on daikon radish.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org