aiqbench analysis
This research introduces Teleoperation Augmentation Primitives (TAPs), which allow operators to trigger specific robot actions via speech or buttons during demonstration, enabling policies to learn capabilities beyond human physical limits. The study demonstrates that these augmentation primitives significantly improve success rates on complex tasks like multi-turn wrist motions and pose anchoring without degrading standard imitation learning performance. By recording these triggers as part of the dataset, the resulting AI policy can autonomously invoke them to solve problems that would otherwise be impossible from visual input alone.
Hybrid Imitation Learning: Teleoperation Augmentation Primitives that Policies Learn to Trigger
arXiv:2512.04960v2 Announce Type: replace Abstract: What an operator can demonstrate bounds what imitation learning can learn. Teleoperation interfaces map the human body to the robot, so motions that are hard for a human, such as holding an exact orientation, returning to the same viewpoint, or turning a wrist joint several full revolutions, are difficult to demonstrate on most teleoperation interfaces, even when they are trivial for the robot. We introduce Teleoperation Augmentation Primitives (TAPs): axis locks, perching waypoints, pose anchors, and embodiment-specific routines that the operator triggers during a demonstration by speech, an AR menu, or simply with a controller button. TAPs are themselves recorded in the demonstration and can therefore also be learned and invoked by the policy itself. In simulation, where benchmarks provide human demonstrations, we show that augmenting them with TAPs after the fact yields better policies on a wristcamera-only peg insertion (0.372 to 0.531 success) and on a mug-cleanup task with a learned "remember this pose" anchor (0.203 to 0.323). On a real robot, three tasks show the same pattern: primitives that help the operator do not hurt the policy, and a policy triggering an unscrewing routine succeeds where a plain policy cannot resolve the multi-turn motion from images alone (67% vs. 38% success). We close with an industrial proof-of-concept case leveraging this idea. Project website: https://hybrid-imitation.github.io/
Abstract:What an operator can demonstrate bounds what imitation learning can learn. Teleoperation interfaces map the human body to the robot, so motions that are hard for a human, such as holding an exact orientation, returning to the same viewpoint, or turning a wrist joint several full revolutions, are difficult to demonstrate on most teleoperation interfaces, even when they are trivial for the robot. We introduce Teleoperation Augmentation Primitives (TAPs): axis locks, perching waypoints, pose anchors, and embodiment-specific routines that the operator triggers during a demonstration by speech, an AR menu, or simply with a controller button. TAPs are themselves recorded in the demonstration and can therefore also be learned and invoked by the policy itself. In simulation, where benchmarks provide human demonstrations, we show that augmenting them with TAPs after the fact yields better policies on a wristcamera-only peg insertion (0.372 to 0.531 success) and on a mug-cleanup task with a learned "remember this pose" anchor (0.203 to 0.323). On a real robot, three tasks show the same pattern: primitives that help the operator do not hurt the policy, and a policy triggering an unscrewing routine succeeds where a plain policy cannot resolve the multi-turn motion from images alone (67% vs. 38% success). We close with an industrial proof-of-concept case leveraging this idea. Project website: this https URL
Robotics (cs.RO)
arXiv:2512.04960 [cs.RO]
(or arXiv:2512.04960v2 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2512.04960 Focus to learn more arXiv-issued DOI via DataCite
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 · 309 words
Related coverage
LadderMan: Learning Humanoid Perceptive Ladder Climbing
arXiv · Jun 6, 2026
Learning Reflexive Behavior for Contact-Rich Manipulation
arXiv · Oct 5, 2026
Wrench-ACT: Enhancing Robot Policies for Contact Rich Behavior Using Direct Wrench Control
arXiv · Sep 30, 2026
WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation
arXiv · Sep 9, 2026