Summary Video
A short overview of the benchmark and its realistic laboratory environments.
Dynamic Experimental Phenomena
Vapor
Flame
Bubble
Parallel Simulation
Grasp Synthesis and Parallel Data Collection
Task Visualizations
Use a pipette to aspirate liquid from a container.
Use tweezers to grasp and manipulate a small object.
Rotate a tube cap to tighten it onto the tube.
Rotate a tube cap to loosen it from the tube.
Attach a disposable tip to a pipette.
Place a tube into a centrifuge slot.
Use the Shadow hand to transfer a tube to its target location.
Use the XHand to transfer a tube to its target location.
Operate the controls on a laboratory mixer panel.
Ignite a Bunsen burner to produce a flame.
Close the lid of a centrifuge.
Heat and stir an agar solution.
Close the lid of a thermal cycler.
Open the lid of a thermal cycler.
Simulation Results
Real-World Results
Both videos show real-world execution of the same base policy(pi 0.5) fine-tuned using simulation data with different visual backgrounds
Gaussian Background vs. MuJoCo Background
The two conditions differ only in the visual background used for the simulated demonstrations.
Fine-tuned on Sim Data with 3DGS Backgrounds
Fine-tuned on Sim Data with MuJoCo Backgrounds
Abstract
Automating biological laboratories requires robots to perform dexterous tool interactions under realistic visual conditions while handling dynamic experimental phenomena, presenting significant challenges for robotic manipulation. To enable systematic evaluation under these conditions, we introduce AutoBio+, a simulation benchmark and framework for dexterous manipulation in realistic biological laboratory environments. AutoBio+ comprises 14 tasks across 4 robot embodiments, spanning functional tool use, precision manipulation, language-conditioned visual grounding, and instrument operation. We introduce controllable dynamic experimental phenomena, including flames, vapor, and bubbles, to capture the evolving visual conditions of laboratory procedures. To further enhance environmental realism, we develop a hybrid rendering pipeline that integrates interactive simulated foreground objects with Gaussian-splatting backgrounds reconstructed from 6 real laboratory scenes. For efficient large-scale data generation and evaluation, we further build a parallel simulation framework on top of mjlab, proposing Parallel Finite-State Machine (PFSM) and integrating batched rendering with decoupled trajectory replay for demonstration generation and model-independent closed-loop policy evaluation. Our evaluation reveals substantial variation in policy performance across tasks, exposing distinct strengths and weaknesses of current manipulation policies. Real-world experiments further demonstrate a reduced sim-to-real gap, supporting the validity of AutoBio+ for realistic manipulation evaluation.