Responsibly advancing AI and robotics
As we explore the continuing potential of AI and robotics, we’re taking a layered, holistic approach to addressing safety in our research, from low-level motor control to high-level semantic understanding.
The physical safety of robots and the people around them is a longstanding, foundational concern within the science of robotics. That is why roboticists have classic safety measures equivalent to avoiding collisions, limiting the magnitude of contact forces, and ensuring the dynamic stability of mobile robots. Gemini Robotics-ER could be interfaced with these ‘low-level’ safety-critical controllers, specific to every particular embodiment. Constructing on Gemini’s core safety features, we enable Gemini Robotics-ER models to grasp whether or not a possible motion is secure to perform in a given context, and to generate appropriate responses.
To advance robotics safety research across academia and industry, we’re also releasing a brand new dataset to judge and improve semantic safety in embodied AI and robotics. In previous work, we showed how a Robot Structure inspired by Isaac Asimov’s Three Laws of Robotics could help prompt an LLM to pick out safer tasks for robots. We have now since developed a framework to mechanically generate data-driven constitutions – rules expressed directly in natural language – to steer a robot’s behavior. This framework would allow people to create, modify and apply constitutions to develop robots which can be safer and more aligned with human values. Finally, the latest ASIMOV dataset will help researchers to carefully measure the security implications of robotic actions in real-world scenarios.
To further assess the societal implications of our work, we collaborate with experts in our Responsible Development and Innovation team and in addition to our Responsibility and Safety Council, an internal review group committed to make sure we develop AI applications responsibly. We also seek the advice of with external specialists on particular challenges and opportunities presented by embodied AI in robotics applications.
Along with our partnership with Apptronik, our Gemini Robotics-ER model can also be available to trusted testers including Agile Robots, Agility Robots, Boston Dynamics, and Enchanted Tools. We look ahead to exploring our models’ capabilities and continuing to develop AI for the subsequent generation of more helpful robots.
