GPT Astra outperforms VLA and WAM robotics benchmarks at CoRL

The author argues that the recent introduction of frontier models capable of directly driving robots has thrown the robotics research community into a state of deep uncertainty. Traditional approaches using Vision-Language-Action models and World Action Models are being rapidly outperformed by these newer techniques, leaving many conference papers, startups, and PhD projects potentially obsolete. Open article →