AI News
AI News AgentModel releaseGoogle2 min read

Gemini Robotics-ER 1.6 improves robotic AI

Google DeepMind introduces Gemini Robotics-ER 1.6, a model that improves spatial understanding, planning and task verification in robots. It can also read complex instruments and is already available to developers through the Gemini API and Google AI Studio.

Google DeepMind has introduced Gemini Robotics-ER 1.6, an artificial intelligence model designed to help robots better understand what is happening around them and act more autonomously during physical tasks.

The update focuses on spatial reasoning. The model can analyze an environment from multiple viewpoints, interpret the position of objects and plan the steps needed to complete a task. It can also check whether the objective was completed correctly.

That makes it possible to move from general instructions to more specific actions. For example, a robot could identify a tool on a table, calculate how to approach it without hitting other objects and then verify whether it placed the tool in the indicated location.

It can also read instruments

One of the new capabilities is reading complex instruments, such as meters, indicators and level gauges, known in industrial settings as sight glasses. The robot can interpret the visual information from these devices to make decisions during a task.

Google DeepMind attributes this capability to a collaboration with Boston Dynamics. In practice, it could allow a robot to check an indicator, detect a reading outside the expected range and act according to the instructions it received. The model does not, on its own, turn any robot into an autonomous machine.

The system brings together several functions relevant to robotics:

  • Visual and spatial understanding of the environment.
  • Step-by-step task planning.
  • Scene analysis from different perspectives.
  • Detection of whether a task has been completed successfully.
  • Reading meters and visual instruments.

Available to developers

Gemini Robotics-ER 1.6 is already available to developers through the Gemini API and Google AI Studio. This makes it possible to test the model and connect it to applications or robotic systems that can receive its instructions and carry out actions in the physical world.

Google also says it is the company's safest robotics model to date. According to the company, it achieved better results in following its safety policies in tests designed to challenge its spatial reasoning with adverse situations. This describes internal evaluation results, not a safety guarantee for every robot or environment.

For you, the most important change is not that robots will immediately start appearing in homes. It is that developers now have a tool better prepared to connect perception, planning and action. The next step will be seeing how it performs outside controlled demonstrations, with unexpected obstacles, difficult-to-read instruments and tasks where a mistake has real consequences.

Gemini Robotics-ER 1.6 improves robotic AI | neversleep.ai