aiqbench analysis
Zero2Skill introduces a novel framework for robot learning that leverages Large Language Models to create a persistent "Corrective Memory," allowing robots to retain human corrections across multiple training sessions rather than repeating fixes. This approach significantly reduces the overhead of autonomous data collection, enabling policies trained on this synthesized data to match teleoperation success rates with only 16% of the traditional human effort required for dataset generation. The method represents a critical advancement in scaling real-world manipulation tasks by decoupling oversight costs from session length and focusing them instead on distinct failure modes.
Zero2Skill: Bootstrapping Robot Skills through Autonomous Data Collection, Training, and Deployment
arXiv:2607.14047v3 Announce Type: replace Abstract: Autonomous data collection governs the volume and quality of real-world trajectories for manipulation policy learning. Existing pipelines reduce human effort via self-resetting, VLM verification, or language-guided correction, yet episode-scoped fixes must be reissued whenever the same failure recurs, so oversight cost grows with session length rather than with the number of distinct problems. We present Zero2Skill, a human-robot symbiotic agentic system in which corrections are retained and reused across rounds. The collection loop collects, verifies, and resets autonomously, pausing for a remote operator only when a phase exhausts an explicit retry budget. An LLM parser maps each natural-language utterance to a structured adjustment stored in Corrective Memory, so addressed failure modes typically need not be corrected again under the same conditions. On a real-robot desktop-clearing testbed, Zero2Skill matches teleoperation episode success while reducing human working time to 16%. Language corrections improve verifier-human agreement in all four evaluated settings and raise average single-attempt success from 12.5% to 47.5% (arm-selection: 20.0% to 50.0%). Policies fine-tuned on Zero2Skill data match teleoperation-trained policy success at a fraction of collection human cost.
Abstract:Autonomous data collection governs the volume and quality of real-world trajectories for manipulation policy learning. Existing pipelines reduce human effort via self-resetting, VLM verification, or language-guided correction, yet episode-scoped fixes must be reissued whenever the same failure recurs, so oversight cost grows with session length rather than with the number of distinct problems. We present Zero2Skill, a human-robot symbiotic agentic system in which corrections are retained and reused across rounds. The collection loop collects, verifies, and resets autonomously, pausing for a remote operator only when a phase exhausts an explicit retry budget. An LLM parser maps each natural-language utterance to a structured adjustment stored in Corrective Memory, so addressed failure modes typically need not be corrected again under the same conditions. On a real-robot desktop-clearing testbed, Zero2Skill matches teleoperation episode success while reducing human working time to 16%. Language corrections improve verifier-human agreement in all four evaluated settings and raise average single-attempt success from 12.5% to 47.5% (arm-selection: 20.0% to 50.0%). Policies fine-tuned on Zero2Skill data match teleoperation-trained policy success at a fraction of collection human cost.
Robotics (cs.RO); Human-Computer Interaction (cs.HC); Systems and Control (eess.SY)
arXiv:2607.14047 [cs.RO]
(or arXiv:2607.14047v3 [cs.RO] for this version)
Full text under the publisher's open licence · 194 words
Related coverage
Practice Makes Policies: Bootstrapping and Consolidating Robotic Capabilities from Zero Human Demonstrations
arXiv · Jul 30, 2026
The Machines that Make the Machines
NVIDIA · Oct 7, 2026
Import AI 471: Why Hugging Face worries me; space mining; FIve Eyes on AI
Import AI · Aug 31, 2026
Problems of Proliferation
LessWrong · Sep 29, 2026