analiza aiqbench
This research introduces WARP-RM, a self-supervised reward model that extracts dense progress signals from noisy robot teleoperation data using time-warp augmentations. By filtering suboptimal action chunks during behavior cloning, the method significantly improves throughput on complex tasks like folding T-shirts and bottle placement without requiring costly human annotations for subtask boundaries. The work advances imitation learning by enabling more robust training from mixed-quality demonstrations, a critical step toward scaling robot autonomy.
WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation
arXiv:2606.28320v4 Announce Type: replace Abstract: Scaling imitation learning requires large datasets, yet human teleoperation inevitably produces mixed-quality demonstrations containing hesitations, retries, and pauses. Prior frame-level progress reward models supervise on absolute temporal progress proxies that suffer from label noise, or require costly human annotations to define subtask boundaries. We present WARP (Warp-Augmented Relative Progress), a novel fully self-supervised algorithm for learning dense, signed relative progress magnitudes directly from successful demonstrations. WARP generates per-frame progress targets via time-warp augmentations of demonstrations (variable playback speeds and reversals) and we train WARP-RM to predict normalized signed temporal displacement from the start of each sampled window. Aggregating these predictions across overlapping windows yields a dense frame-level progress signal. We then introduce WARP-BC, which uses these scalar reward estimates to filter and reweight action chunks during behavior cloning. We evaluate our approach on a physical bimanual robot system performing a long-horizon deformable object manipulation task: folding T-shirts from a random crumpled start. To evaluate policy robustness against suboptimal data, we construct training datasets of varying quality using episode length as a proxy for teleoperation sub-optimality. Across these training tiers, WARP-BC improves successful-folding throughput by up to ~18x over vanilla BC. Furthermore, we evaluate bottle-in-bin placement in the real world and in simulation. Across 512 paired simulated scenes, WARP-BC achieves 290 bottles/hr versus 237 for vanilla BC and 271 for DemInf, with all curation methods retaining 31.5% of the data. We release open simulation data, code, checkpoints, and evaluation artifacts for end-to-end reproduction of the WARP pipeline. Project page: https://uynitsuj.github.io/warp-rm/
Abstract:Scaling imitation learning requires large datasets, yet human teleoperation inevitably produces mixed-quality demonstrations containing hesitations, retries, and pauses. Prior frame-level progress reward models supervise on absolute temporal progress proxies that suffer from label noise, or require costly human annotations to define subtask boundaries. We present WARP (Warp-Augmented Relative Progress), a novel fully self-supervised algorithm for learning dense, signed relative progress magnitudes directly from successful demonstrations. WARP generates per-frame progress targets via time-warp augmentations of demonstrations (variable playback speeds and reversals) and we train WARP-RM to predict normalized signed temporal displacement from the start of each sampled window. Aggregating these predictions across overlapping windows yields a dense frame-level progress signal. We then introduce WARP-BC, which uses these scalar reward estimates to filter and reweight action chunks during behavior cloning. We evaluate our approach on a physical bimanual robot system performing a long-horizon deformable object manipulation task: folding T-shirts from a random crumpled start. To evaluate policy robustness against suboptimal data, we construct training datasets of varying quality using episode length as a proxy for teleoperation sub-optimality. Across these training tiers, WARP-BC improves successful-folding throughput by up to ~18x over vanilla BC. Furthermore, we evaluate bottle-in-bin placement in the real world and in simulation. Across 512 paired simulated scenes, WARP-BC achieves 290 bottles/hr versus 237 for vanilla BC and 271 for DemInf, with all curation methods retaining 31.5% of the data. We release open simulation data, code, checkpoints, and evaluation artifacts for end-to-end reproduction of the WARP pipeline. Project page: this https URL
Robotics (cs.RO)
arXiv:2606.28320 [cs.RO]
(or arXiv:2606.28320v4 [cs.RO] for this version)
https://doi.org/10.48550/arXiv.2606.28320 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.
Text integral sub licența deschisă a publicației · 342 cuvinte
Subiecte conexe
Expenditure Horizon: Measuring Optimization Ability, with an Application to NanoGPT
METR · 21 iul. 2026
Fable is SOTA at CIFAR Speedrun (& specification gaming)
LessWrong · 20 iul. 2026
Problems of Proliferation
LessWrong · 29 sept. 2026
What's Continual Learning, and Why Might We Expect To See It In Advanced LLM Agents?
LessWrong · 12 iun. 2026