CoDA: Co-Evolution of Data Engines and Action Policies
arXiv cs.AI · 16 Oct 2026, 03:59 AEST
CoDA is a framework that enables the co-evolution of simulation data engines and action policies to improve robot training environments. It enhances out-of-distribution success rates significantly across various manipulation tasks.
Key facts
- Generative simulation can automate robot environment construction, but physically valid environments are not necessarily useful for training.
- We introduce CoDA, a framework in which a simulation data engine and an action policy co-evolve through a shared heterogeneous memory of environment construction, policy failures, and learning progress.
- Across four manipulation task families, CoDA improves out-of-distribution success from 5.3% after supervised fine-tuning to 27.6%, exceeding the strongest closed-loop baseline by 9.5 percentage points with the same generation budget.
- On LIBERO-Pro, 15 rounds of generation and training raise success from 25% to 84% under position perturbations and from 1% to 76% under instruction perturbations.
Read aloud
CoDA is a framework that enables the co-evolution of simulation data engines and action policies to improve robot training environments. It enhances out-of-distribution success rates significantly across various manipulation tasks. Generative simulation can automate robot environment construction, but physically valid environments are not necessarily useful for training. We introduce CoDA, a framework in which a simulation data engine and an action policy co-evolve through a shared heterogeneous memory of environment construction, policy failures, and learning progress. Across four manipulation task families, CoDA improves out-of-distribution success from five point three percent after supervised fine-tuning to twenty seven point six percent, exceeding the strongest closed-loop baseline by nine point five percentage points with the same generation budget. On LIBERO-Pro, fifteen rounds of generation and training raise success from twenty five percent to eighty four percent under position perturbations and from one percent to seventy six percent under instruction perturbations.
Sources in this cluster
- CoDA: Co-Evolution of Data Engines and Action Policies
arXiv cs.AI · 16 Oct 2026, 03:59 AEST