Event

RoboGene: A Task Generation Framework for Multi-Morphology Robot VLA Pretraining

1 reports · 1 sources

Yixue Zhang and other authors released a paper proposing the RoboGene framework, used before VLA pretraining to automatically generate diverse manipulation tasks for single-arm, dual-arm, and mobile robots. The framework combines diversity-driven sampling, agent self-reflection to evaluate task quality, and a human-verified feedback memory for execution validation and continuous improvement, collecting a total of 18,000 trajectories and introducing task feasibility and diversity metrics. The authors state that RoboGene significantly outperforms frontier foundation models such as GPT-4o and Gemini 2.5 Pro in task generation; VLA models trained on data generated by RoboGene achieve higher success rates and stronger generalization ability. The above comparisons and effects are all claims made by the authors in the paper.