Pipette: An Embodied Simulation Platform, Benchmark, and Data-Efficient Augmentation Framework for Wet-Lab Robotics

Jul 16, 2026

Upload by Aurora

Wet-lab robots enhance reproducibility, throughput, and safety in biomedical experiments, yet scaling learning demands customizable simulators for safe task generation, openly editable lab assets, and efficient pipelines that convert limited demonstrations into usable training data. Pipette introduces an embodied simulation platform, benchmark, and data-efficient augmentation framework for wet-lab robot learning, delivering 100+ open re-editable assets, unified multi-arm simulation, and simulation replay with perturbations and automatic success filtering. A 12-task wet-lab benchmark validates improved ACT and SmolVLA/pπ0 performance, while natural-language scene construction lowers entry barriers for new tasks.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages19

Reading Time48 min

Abstract

Introduction

Challenges in Biomedical Wet-Lab Robot Learning

Pipette Platform and Contributions

Pipette: An Embodied Simulation Platform, Benchmark, and Data-Efficient Augmentation Framework for Wet-Lab Robotics Zhe Liu∗1,2, Huanbo Jin∗1,2, Zhaohui Du∗1,2, Zhe Wang†1,2, Dongzhan Zhou4 , Minting Pan4 , He Xu2 , Peijia Li2 , Jiaming Gu 1,2 , Quan Lu 1,2 , Qi Wang3 , Bin Ji 1,2 , Ting Xiao 1,2 1 Key Laboratory of Smart Manufacturing in Energy Chemical Process, Ministry of Education, East China University of Science and Technology, Shanghai, China 2Department of Computer Science and Engineering, East China University of Science and Technology, Shanghai, China 3Department of Laboratory Medicine, Ruijin Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, China 4AI for Science Center, Shanghai AI Laboratory, Shanghai, CN [wangzhe@ecust.edu.cn](wangzhe@ecust.edu.cn) arXiv :2606 . 12936v2 [ cs .RO] 13 Jul 2026 Abstract Wet-lab robots can improve the reproducibility, throughput, and safety of biomedical experiments, but scaling their learning requires customizable simulators for safe and reproducible task generation, open editable laboratory assets, and efficient pipelines that turn limited demonstrations into usable training data. We present Pipette, an embodied simulation platform, benchmark, and data-efficient augmentation framework for wet-lab robot learning. Pipette provides over 100 open-source and re-editable wet-lab assets through an extensible assetbuilding pipeline with built-in Tencent Hunyuan support for text- and image-conditioned 3D asset generation, and supports three robotic-arm embodiments through a unified simulation interface for task construction, data collection, augmentation, and evaluation. A key component of Pipette is its simulation-based data augmentation pipeline, which replays human demonstrations in simulation, applies lighting, camera, speed, and action perturbations, and filters generated episodes with automatic task success checks, rapidly expanding usable training data from limited manual demonstrations. We further introduce a 12-task wet-lab embodied benchmark covering sample handling, culture-ware manipulation, device operation, and precision placement. With only 30 demonstrations per task, ACT achieves 60.3% average success rate, while simulation augmentation improves SmolVLA from 40.4% to 71.8% and π0 from 37.3% to 44.1%, validating the effectiveness of Pipette for data-efficient VLA training and evaluation. Pipette also supports natural-language-driven scene construction and task registration, lowering the barrier for non-expert users to define new wet-lab robotic tasks. 1 Introduction Biomedical laboratory automation is increasingly important for reproducible, efficient, and safe life science experi- ∗These authors contributed equally. †Corresponding author. Copyright © 2027, Association for the Advancement of Artificial Intelligence ([www.aaai.org](www.aaai.org)). All rights reserved. ments. Automated systems reduce manual variability in highthroughput processing, standardized protocols, and longterm repetitive operations (Holland and Davies 2020), but complex protocols still stress timing, consistency, and system integration (Rupp et al. 2024) . Self-driving laboratories combine AI, instruments, and automated pipelines for experimental design, execution, analysis, and optimization (Tobias and Wahab 2025), while agentic AI extends automation toward task planning and closed-loop experimentation (Hartung 2025) . However, many biomedical systems remain tied to fixed instruments, scripts, and task-specific workflows, making them hard to adapt to transparent consumables, small targets, interactive devices, and multi-step wet-lab manipulation. Scalable wet-lab automation therefore needs an embodied environment that connects protocols, visual perception, and robotic action. Recent robot foundation models and vision-languageaction (VLA) models suggest a path toward general-purpose control. RT-1 and RT-2 show that large-scale robot data and we

FAQs

What problem does Pipette address in wet-lab robot learning?

Pipette targets the need for safe, reproducible task generation and scalable simulation support, because collecting demonstrations in biomedical wet labs is costly and constrained by safety, fragility, and contamination risks.

What does Pipette provide to enable training and evaluation?

Pipette offers an embodied simulation platform with an extensible asset-building pipeline, a unified simulation interface across three robotic-arm embodiments, and tools for task construction, data collection, augmentation, and evaluation.

How does Pipette improve performance using limited demonstrations?

It uses a simulation-based data augmentation pipeline that replays human demonstrations, applies controlled perturbations (lighting, camera, speed, actions), and filters episodes with automatic task-success checks, expanding training data efficiently.

Jul 16, 2026

Upload by Aurora