Gemini Robotics 1.5: robots can now ‘think’!

Gemini Robotics 1.5 has introduced a groundbreaking agentic framework that empowers robots to autonomously tackle complex, multi-step physical tasks with unprecedented intelligence and dexterity.

Also read: China’s AI-powered humanoid robots aim to transform manufacturing

Everyday activities, like sorting recyclables based on local guidelines or sorting laundry by color etc. pose significant challenges for robots. Well they did at least, until now! These tasks require contextual reasoning, multi-step planning, and real-time adaptation. Gemini Robotics 1.5 has come up with robots that meet these demands.

For those familiar with the techie details, these robots perform such functions by combining two specialised models working in tandem:

• Gemini Robotics-ER 1.5, an embodied reasoning “brain” that excels at high-level planning, spatial understanding, and tool use. It is capable of querying live data sources like Google Search and estimating task progress dynamically.
• Gemini Robotics 1.5, a vision-language-action model that executes each task step, interpreting natural language instructions and explaining its decision-making process, enhancing transparency and semantic reasoning.

Together, these models unlock agentic experiences where robots don’t just follow commands but think before acting, breaking down complex instructions into manageable subtasks while adapting to new environments and diverse robot embodiments without retraining.

Also read: Humanoid robots to aid, not replace, humans, says Chinese official

Gemini Robotics-ER 1.5 dominates 15 academic embodied reasoning benchmarks, setting new state-of-the-art performance in tasks from object pointing to video question answering. Meanwhile, the system’s robust safety architecture integrates semantic reasoning and real-time collision avoidance, aligning with responsible AI principles and raising the bar for trustworthy robotics.

This advancement heralds a critical milestone toward Artificial General Intelligence (AGI) in the physical world. Now we have robots that not only react but reason, plan, and autonomously navigate our complex environments.