Beijing, China, Oct. 29, 2026 (GLOBE NEWSWIRE) -- PokeBot, an embodied AI company focused on developing intelligent systems that can understand and act in the physical world, today announced the launch of RoboAPI, an interface that gives developers access to Poke-Aurora, the company’s world-action model, for testing on compatible robots.
RoboAPI is designed to make embodied AI easier to evaluate on physical robotic systems without requiring developers to train a new model from scratch. By providing access to Poke-Aurora through a single interface, RoboAPI aims to reduce the need for additional data collection and model tuning, giving developers a way to explore the model’s capabilities in their own environments and across different tasks.
Bringing Embodied AI Into the Physical WorldUnlike AI systems that operate primarily in digital environments, embodied AI must interpret information from the physical world and translate it into actions.Developing these capabilities often requires extensive data collection, model training, and hardware-specific adaptation, creating challenges for researchers and developers working with different robotic platforms.
RoboAPI is designed to lower these barriers by enabling developers to connect compatible robots to Poke-Aurora. Through the interface, the model can process observations, interpret instructions, and generate predicted robot actions, allowing developers to evaluate its performance on physical hardware.Rather than building and training a new model for each application, developers can use RoboAPI to explore Poke-Aurora’s capabilities across tasks and environments. This approach is intended to make real-world testing more accessible and help developers better understand how embodied AI models perform beyond controlled demonstrations.
Introducing Poke-Aurora: A World-Action ModelPoke-Aurora is PokeBot’s world-action model, accessible to developers through RoboAPI. The model is designed to support generalization across tasks and environments, including situations that differ from those encountered during training.By connecting Poke-Aurora to compatible robots through RoboAPI, developers can explore how the model responds to instructions and changing real-world conditions.
In PokeBot’s large-scale evaluation covering nearly 1,000 unseen objects and nearly 10,000 unseen tasks, Poke-Aurora achieved an overall task success rate of 74.2%, surpassing pi0.5 by 51.3% and GPT-6 Astra by 16.4%.
“Building capable robots is only part of the challenge,” said Huazhe Xu, founder of PokeBot and an assistant professor at Tsinghua University’s Institute for Interdisciplinary Information Sciences.
“We also need to make it easier for developers to put embodied AI to the test in the real world,” said Xu. “With RoboAPI, we’re opening access to Poke-Aurora so developers can explore how it performs on compatible robots and different tasks. We want to make it easier to move from seeing what these models can do in a demo to understanding how they work in practice.”
Technical Paper and Product Demo Available Oct. 30PokeBot will release the technical paper and product demonstration for RoboAPI on Oct. 30, 2026, Beijing time, alongside the launch. RoboAPI gives developers access to Poke-Aurora, PokeBot’s world-action model, for testing on compatible robots. During the technical preview, RoboAPI will be available as a noncommercial offering to eligible users.
About PokeBotPokeBot is an embodied AI company developing intelligent systems that enable robots to perceive, understand, and act in the physical world. Through its work on general-purpose embodied AI models, the company aims to advance adaptable robot intelligence and bring capable robots closer to everyday applications.Learn more at pokebot.com.
