RoboDrop improves robot learning by filtering training data errors
RoboDrop: Curating VLA Post-Training Data via Local Gradient Compatibility
Robotics
Summary
Robots learn new tasks by training on data collected from their own actions, but this data often has mistakes like sensor errors or timing problems that confuse the learning process. The authors present RoboDrop, a tool that automatically checks training data for quality by watching how each example influences learning progress. It scores and filters out bad data, helping the robot learn more effectively. Testing shows RoboDrop improves the robot’s success rate by removing problematic examples from the training set.
What this means in practice
- •For robotics engineers: Improve new task training by automatically filtering out bad robot demonstration data to boost policy success rates.
- •For industrial automation teams: Enhance robotic system reliability by auditing collected training data to detect execution errors and sensor faults before retraining.
Authors
Runze Xu, Yuanfan Xu, Cuijie Xu, Shuang Dai, Yining Li, Yu Wang, Jincheng Yu
Abstract
Vision--language--action (VLA) models acquire broad generalization through large-scale pretraining, yet adapting them to a new task and robot embodiment still requires post-training on newly collected data. Unlike pretraining, post-training targets task- and embodiment-specific adaptation, making it particularly sensitive to data quality. In practice, collected robot datasets often contain heterogeneous errors, including execution mistakes, sensor drift, and timestamp misalignment, which can impair post-training and policy performance. Manual inspection is costly, while existing data-cleaning methods are typically tailored to particular corruption types. To address these challenges, we introduce \textsc{RoboDrop}, a data-curation framework that audits supervision using local gradient compatibility measured along the training trajectory as a proxy for its effect on post-training performance. During a one-epoch warm-up run, RoboDrop scores each candidate sample online by comparing its gradient with those of task-semantic and visually matched validation samples. The resulting sample scores are aggregated at the episode level, and a simple automatic post-processing rule converts them into filtering decisions. We evaluate RoboDrop on controlled observation--action corruptions, naturally suboptimal demonstrations in simulation, and real-robot datasets containing non-expert collection errors. Across these settings, RoboDrop more accurately distinguishes unreliable demonstrations than prior methods, while post-training on the curated data consistently yields stronger downstream policies, with average real-robot rollout success rising from $35.0\%$ to $67.5\%$. These results establish training-trajectory-aware, context-conditioned supervision auditing as an effective approach to robust VLA post-training.