How AI and LLMs Are Reshaping Robotics

Author(s): Adit Sheth

Originally published on Towards AI.

For years, the promise of robots seamlessly integrated into our daily lives and industries felt like a distant science fiction dream. Yet, in the blink of an eye, the rapid advancements in Artificial Intelligence (AI) and Large Language Models (LLMs) are transforming this vision into a tangible reality. The last two weeks (roughly mid-June to early July 2025) have been particularly transformative, showcasing a future where AI isn’t just confined to screens but is actively assisting, reasoning, and adapting in the physical world.

This isn’t merely about technological marvel; it’s about how these groundbreaking innovations are poised to impact diverse communities — from factory workers and logistics professionals to everyday consumers. Let’s delve into some of the most compelling recent developments in AI robotics.

I. Google’s Gemini Robotics On-Device: Unleashing True Robot Autonomy

One of the most significant and exciting breakthroughs to emerge recently is Google DeepMind’s Gemini Robotics On-Device. Launched in late June 2025, this innovation represents a pivotal shift in how robots will operate. For the first time, highly sophisticated multimodal vision-language-action (VLA) models can run directly on robotic platforms, entirely offline, eliminating the need for constant internet connectivity or cloud inference.

Previously, many advanced robotic AI systems relied on sending data to powerful cloud servers for processing. While effective, this approach introduced inherent limitations: latency (delays in response), reliance on stable internet, and potential data privacy concerns, especially in sensitive environments like homes or hospitals. Gemini Robotics On-Device fundamentally changes this paradigm.

Optimized for low-compute hardware, this on-device model retains the advanced capabilities typically associated with larger, cloud-based LLMs. Robots equipped with this technology can now perform:

• Natural Language Understanding: Interpreting complex voice commands like “Please hand me the blue wrench from the top shelf.”
• Multi-Step Task Execution: Breaking down an overarching instruction into a sequence of smaller, manageable actions.
• Complex Manipulation: Handling delicate objects and performing intricate tasks with remarkable precision, as demonstrated by examples like folding clothes, unzipping bags, and assembling mechanical parts.

The immediate benefits are profound:

• Faster Response Times: Robots can react to their environment and human commands almost instantaneously, crucial for safety and efficiency in dynamic settings.
• Enhanced Data Privacy: Sensitive visual or environmental data can be processed locally, significantly reducing the need to transmit it to external servers, thereby bolstering privacy.
• Improved Reliability: Operations become less vulnerable to network outages or slowdowns, making robots more dependable in critical applications and remote locations.

Google DeepMind has made an SDK for Gemini Robotics On-Device available to trusted testers, empowering a wider community of developers to build and deploy sophisticated robotic applications. This ability to run powerful AI directly on the robot, coupled with rapid adaptation capabilities (requiring as few as 50 to 100 demonstrations to fine-tune for new tasks), represents a monumental step toward truly autonomous and responsive robots.

Innovation for Community: The implications span across numerous sectors. For manufacturing and logistics, autonomous robots can now navigate complex warehouses and production lines with unprecedented efficiency and reduced reliance on centralized control. In healthcare, assistive robots could perform sensitive patient care tasks with enhanced privacy and real-time responsiveness. For consumers, this technology paves the way for more intelligent, responsive, and truly helpful robots in homes, performing tasks previously unimaginable for a self-contained unit. It shifts the paradigm towards truly autonomous, adaptable, and internet-independent robots, a pivotal leap for the future of physical AI.

II. Humanoid Robots Enter the Factory: Foxconn and Nvidia’s Bold Move

While Gemini Robotics On-Device pushes the boundaries of individual robot intelligence, another major recent announcement points to a broader, industry-wide shift: Foxconn and Nvidia’s plans to deploy humanoid robots on AI server production lines. Reports from late June 2025 indicate that these human-like robots could begin assisting in assembly at Foxconn’s new Houston factory as early as the first quarter of 2026.

This collaboration is a significant milestone for several reasons:

• First-Ever Scale Deployment: This marks one of the earliest planned integrations of humanoid robots into a high-tech manufacturing process at such a significant scale, especially for the production of critical AI infrastructure like Nvidia’s next-generation GB300 AI servers.
• Complex Task Handling: Foxconn has reportedly been training these robots for tasks such as item handling, cable plugging, and assembly — jobs that require a high degree of dexterity and precision traditionally performed by human workers.
• Nvidia’s Robotics Platform: Nvidia, a leading provider of AI computing, is also a major player in robotics, providing powerful platforms like NVIDIA Isaac GR00T N1.5, an open foundation model for generalized humanoid robot reasoning and skills. This partnership strategically leverages their expertise to accelerate the deployment of intelligent robots in a demanding industrial environment. GR00T N1.5 itself, updated in May 2025, has demonstrated vastly improved language following (93.3% success rate on real humanoid robots for specific instructions, up from 46.6% in previous versions), better data efficiency for training, and superior generalization to novel objects and environments.

Innovation for Community: This collaboration signals a tangible acceleration in the adoption of advanced robotics in manufacturing. For businesses, it promises increased productivity, improved safety in hazardous environments, and the ability to scale production more efficiently to meet surging demand for AI hardware. For the workforce, it underscores the evolving nature of jobs in manufacturing, shifting from purely manual labor to roles involving robot oversight, programming, and maintenance. It highlights the growing need for new skills and training programs to prepare workers for this new era of human-robot collaboration, emphasizing that these robots are more likely to augment, rather than entirely replace, human capabilities in the near term.

III. The Broader Landscape: LLMs as the “Brain” of Robots

These recent announcements from Google and the Foxconn-Nvidia partnership are indicative of a larger, profound trend: the integration of LLMs as the cognitive “brain” for robots. LLMs, with their unparalleled ability to understand natural language, generate coherent responses, and process vast amounts of information, are increasingly being used to imbue robots with higher-level intelligence.

This fundamental shift is being driven by innovations across the board, enabling robots to:

• Interpret Complex, Ambiguous Commands: Moving beyond rigid, pre-programmed instructions, LLMs allow users to give robots natural language commands like “Clean up the living room and put everything back where it belongs,” relying on the robot’s AI to interpret “clean up” and “where it belongs” based on its learned understanding of the environment.
• Perform Multi-Step Reasoning and Planning: LLMs can break down abstract goals into smaller, manageable steps, and even adapt their plans in real-time if conditions change (e.g., if a requested object isn’t where it’s expected, the robot can infer a new search strategy). This capability is crucial for robots operating in unpredictable real-world environments.
• Generalize Across Tasks: Unlike older robots that needed extensive re-programming for each new task, LLM-powered robots can leverage their vast knowledge base and reasoning abilities to generalize and adapt to unfamiliar situations with minimal additional training, significantly reducing deployment time and cost for new applications.
• Improve Human-Robot Interaction: LLMs facilitate more natural and intuitive communication, allowing robots to provide clear updates, ask clarifying questions, and even learn preferences through conversation, making human-robot collaboration more seamless and effective. Siemens, for instance, has been advancing autonomous production with new AI and robotics capabilities for automated guided vehicles, enabling more dynamic and flexible factory floors.

Conclusion

This convergence of AI, LLMs, and physical robotics is poised to redefine numerous industries. From augmenting human labor in hazardous environments to providing personalized assistance in homes and healthcare facilities, the potential applications are boundless. The rapid pace of innovation in the past two weeks demonstrates that the era of intelligent, physically capable AI is no longer a distant dream, but a rapidly unfolding reality. As we continue to witness these powerful capabilities emerge, the focus will increasingly shift to how we responsibly integrate them to create a future where humans and intelligent machines can thrive together.

References

Google DeepMind’s Gemini Robotics On-Device Launch

• https://deepmind.google/discover/blog/gemini-robotics-on-device-brings-ai-to-local-robotic-devices/ (Official announcement, June 24, 2025)
• https://roboticsandautomationnews.com/2025/06/26/google-deepmind-launches-new-vision-language-action-model-to-put-ai-directly-into-local-robotic-devices/92669/ (Robotics & Automation News, June 26, 2025)

Foxconn and Nvidia Humanoid Robot Deployment

• https://www.pymnts.com/artificial-intelligence-2/2025/nvidia-and-foxconn-aim-to-use-humanoid-robots-in-ai-server-factory/ (PYMNTS.com, June 20, 2025)
• https://www.tomshardware.com/tech-industry/semiconductors/nvidia-products-could-be-made-using-humanoid-robots-for-the-first-time-ever-company-in-talks-with-foxconn-to-deploy-them-in-houston-factory-building-gb300-ai-servers (Tom’s Hardware, June 20, 2025)

NVIDIA Isaac GR00T N1.5 and Broader Robotics Context

• https://learnopencv.com/gr00t-n1_5-explained/ (LearnOpenCV, June 12, 2025)
• https://blogs.nvidia.com/blog/automatica-robotics-2025/ (NVIDIA Blog, June 24, 2025)

Join thousands of data leaders on the AI newsletter. Join over 80,000 subscribers and keep up to date with the latest developments in AI. From research to projects and ideas. If you are building an AI startup, an AI-related product, or a service, we invite you to consider becoming a sponsor.

Published via Towards AI