aiqbench analysis
FineART introduces a densely annotated dataset for bimanual robotic manipulation, addressing the critical gap in long-horizon task execution where existing single-arm datasets fall short. The accompanying FineART-VLA model demonstrates significant improvements in subtask prediction and zero-shot generalization to unseen hardware through mid-training techniques. This work advances the field by providing essential data infrastructure for training vision-language-action policies capable of complex, multi-step real-world operations.
FineART: Fine-grained Annotated Robotic Trajectory Dataset and Vision-Language-Action Model for Bimanual Manipulation
arXiv:2609.36416v1 Announce Type: new Abstract: Robots operating in real-world environments must execute complex, multi-step bimanual tasks over long horizons rather than single, isolated actions. Current manipulation datasets struggle to support this capability: although single-arm datasets reach hundreds of thousands of trajectories, they typically provide only one high-level instruction per episode while the rare bimanual effort that does label subtasks annotates only a fraction of its hours. We present FineART, a densely annotated bimanual manipulation dataset of 40,543 episodes, 1,718 hours, and 533,913 subtasks across 151 tasks. We also introduce FineART-VLA, a vision-language-action policy that predicts its own next subtask, and show that mid-training it this way yields substantial gains. Specifically, success on a spatial disambiguation task increases from 32.0% to 100.0%, and step-by-step human subtask guidance lifts success on an unseen long-horizon task from 16.0% to 76.0%. Furthermore, after minimal fine-tuning on a new robot, the policy requires one-tenth the data of baselines without mid-training and generalizes zero-shot to completely unseen tasks on the new hardware. We open-source the full dataset, model weights, and training code.
Abstract:Robots operating in real-world environments must execute complex, multi-step bimanual tasks over long horizons rather than single, isolated actions. Current manipulation datasets struggle to support this capability: although single-arm datasets reach hundreds of thousands of trajectories, they typically provide only one high-level instruction per episode while the rare bimanual effort that does label subtasks annotates only a fraction of its hours. We present FineART, a densely annotated bimanual manipulation dataset of 40,543 episodes, 1,718 hours, and 533,913 subtasks across 151 tasks. We also introduce FineART-VLA, a vision-language-action policy that predicts its own next subtask, and show that mid-training it this way yields substantial gains. Specifically, success on a spatial disambiguation task increases from 32.0% to 100.0%, and step-by-step human subtask guidance lifts success on an unseen long-horizon task from 16.0% to 76.0%. Furthermore, after minimal fine-tuning on a new robot, the policy requires one-tenth the data of baselines without mid-training and generalizes zero-shot to completely unseen tasks on the new hardware. We open-source the full dataset, model weights, and training code.
26 pages. Code and model weights will be integrated into Hugging Face LeRobot this https URL
Robotics (cs.RO); Artificial Intelligence (cs.AI); Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2609.36416 [cs.RO]
(or arXiv:2609.36416v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2609.36416 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 292 words
Related coverage
FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies
arXiv · Sep 3, 2026
FineVLA: Fine-Grained Instruction Alignment for Steerable Vision-Language-Action Policies
arXiv · Jun 16, 2026
ExecVLA: Following Fine-Grained Execution Constraints in Vision-Language-Action Models with Bi-Level Action Representation
arXiv · Oct 10, 2026
Think Like a Pilot: Fine-Grained Long-Horizon UAV Navigation
arXiv · Jun 8, 2026