aiqbench analysis
This report shows that a scripted teacher that almost always succeeds can still train a poor imitation-learning student. The student inherits stalls, shortcuts, and no exposure to its own errors, and fixing the teacher to be imitable lifted one task's score from 0.44 to 0.82 with the same small architecture. For ML readers, the main takeaway is that demonstration design and evaluation choices can matter more than model size. The results come from simulated competition tasks, so transfer to real robots remains untested.
Learning from an Imitable Teacher: Competition Solution for the MetaFine Challenge
arXiv:2611.00615v1 Announce Type: cross Abstract: The MetaFine Open Competition (RoboFineMani workshop, CoRL 2026) evaluates fine-grained manipulation beyond binary success. Its five simulated tasks, from grasping a bottle by an instructed part to inserting letter-shaped pegs with 4 mm clearance, are scored on perception under lighting and camera changes, instruction understanding, and behavior. This report presents Team Friday's solution, which got second place with an overall score of 0.947, more than twice the $\pi_0$ baseline. Instead of fine-tuning a large vision-language-action (VLA) model, we train one 19M-parameter policy per task with a separate instruction classifier and focus on the data, all generated in the simulator by a privileged scripted teacher, including its corrections of the student's mistakes. By analyzing failed episodes, we find that a teacher that almost always succeeds can still teach a memoryless student badly: its pauses become stalls, its fixed habits become shortcuts, and it never shows the errors the student makes. Making the teacher \emph{imitable} raised the official score on letter insertion, the hardest task, from 0.44 to 0.82 with 32\% more data and the same architecture, and finally to 0.88. We summarize practical lessons on demonstration data and evaluation, which we hope will help in imitation learning for precise manipulation.
Abstract:The MetaFine Open Competition (RoboFineMani workshop, CoRL 2026) evaluates fine-grained manipulation beyond binary success. Its five simulated tasks, from grasping a bottle by an instructed part to inserting letter-shaped pegs with 4 mm clearance, are scored on perception under lighting and camera changes, instruction understanding, and behavior. This report presents Team Friday's solution, which got second place with an overall score of 0.947, more than twice the $\pi_0$ baseline. Instead of fine-tuning a large vision-language-action (VLA) model, we train one 19M-parameter policy per task with a separate instruction classifier and focus on the data, all generated in the simulator by a privileged scripted teacher, including its corrections of the student's mistakes. By analyzing failed episodes, we find that a teacher that almost always succeeds can still teach a memoryless student badly: its pauses become stalls, its fixed habits become shortcuts, and it never shows the errors the student makes. Making the teacher \emph{imitable} raised the official score on letter insertion, the hardest task, from 0.44 to 0.82 with 32\% more data and the same architecture, and finally to 0.88. We summarize practical lessons on demonstration data and evaluation, which we hope will help in imitation learning for precise manipulation.
Robotics (cs.RO); Artificial Intelligence (cs.AI)
arXiv:2611.00615 [cs.RO]
(or arXiv:2611.00615v1 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2611.00615 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 · 295 words
Related coverage
DEF CON 34 Opens Today: AI Agents Graduate From Novelty to Standard Hacking Weapon
techtimes.com · Aug 6, 2026
Information-Time Proximal Policy Optimization
arXiv · Sep 22, 2026
The Disruptive Impact of Large Language Models on Capture the Flag Competitions and the Path Toward Fair Play
arXiv · Jul 29, 2026
LATTEArena: An Evaluation Framework for LLM-powered Tabular Feature Engineering (Extended Version)
arXiv · Jun 9, 2026