DeepMind Introduces Gemini Robotics 2.5 and Shares First Open‑Source Safety Benchmark

During a briefing, Google's DeepMind unit unveiled the newest version of its Gemini Robotics suite, called Gemini Robotics 2.5, emphasizing that the system is increasingly capable of handling routine chores with competence nearing that of humans.

This release brings a collection of software upgrades that sharpen the link between Gemini’s large‑language‑model core and the physical control loops governing robot arms. DeepMind says the refinements allow a number of prototype machines to carry out real‑world tasks—like sorting conveyor‑belt items, putting together basic parts, and moving through crowded spaces—without any direct human tele‑control.

At the same time as the product launch, DeepMind made public an open safety benchmark named "ASIMOV‑jailbreak." The test is intended to measure how well a robotic AI can withstand prompt‑injection attacks that try to bypass its built‑in safety limits. By sharing the suite openly, DeepMind aims to spur wider community examination of safety features in embodied AI.

The Gemini system builds on prior iterations that fused multimodal perception with language‑driven planning. Whereas earlier versions proved the concept in tidy laboratory environments, the 2.5 update signals a move toward greater autonomy in fluid, unstructured settings. Analysts point out that attaining dependable task performance in these scenarios remains a major obstacle for bringing service robots to market.

DeepMind’s choice to release the ASIMOV‑jailbreak benchmark underscores rising worries about the possible abuse of strong language models housed in physical agents. Scholars have cautioned that hostile prompts might persuade robots to act unsafely, rendering open testing frameworks crucial for responsible progress.

Looking forward, DeepMind said its next steps will involve scaling the platform to tackle more intricate tasks and broadening the benchmark to encompass a larger array of safety situations. This initiative could encourage other AI laboratories to embrace comparable open‑source safety norms, possibly influencing the regulatory framework for robotics as the field advances from prototypes to full‑scale production.