Why GPT-6 Astra is suddenly so good at robotics (twitter.com)

The recent release of GPT-6 Astra by OpenAI has significantly advanced robotics capabilities, prompting discussions in the AI/ML community about its implications for robot programming. In a striking demonstration by researchers at the University of Illinois, Astra was able to write its own code to successfully operate a robot arm to open a drawer, iterating on its programming based on sensor feedback. This marks a pivotal shift away from traditional robotics, where extensive human demonstrations were necessary, as Astra can utilize vision-language-action models to directly convert visual input and instructions into motor commands. The performance of GPT-6 Astra on the RoboDojo benchmarking tasks—outperforming existing robot policies without any specific fine-tuning—illustrates a breakthrough in simplifying the interaction between robot "brains" and their "hands." By employing a more efficient build order that allows the model to write its operational code directly, the robotics industry may be poised for a transformation akin to the introduction of GPT-3 in natural language processing. This change could disrupt the current paradigm where the focus is predominantly on collecting human demonstration data and traditional programming methods, underscoring the potential of AI models like Astra to redefine how robots learn and perform tasks autonomously.