Mbodi AI - A New Paradigm for AI Robotics and Our Partnership with ABB

Tuesday, December 17, 2024

The field of robotics has faced challenges in meeting its full potential for quite some time. A key issue has been the scalability of algorithm-based solutions, which differ significantly from typical SAAS or Internet business models. While companies like Netflix can serve billions of users with relatively stable software, in robotics each new application requires solving a unique, complex problem from the ground up.

Recent advancements in vision language action models (VLAs) show the possibility of generalizability in robotics, which marks a shift in this paradigm. These models show promise in laboratory settings, albeit with significant resource requirements, leading some industry experts to anticipate a "ChatGPT-like" breakthrough where a foundation model revolutionizes the field. At Mbodi, we see the future of robotics as a rapid yet continuous evolution, more akin to the development of PCs or smartphones. We believe that integrating current technologies like (vision) language models, computer vision models, and cloud computing to work together cohesively and improve the models and system over time will lead to comprehensive solutions and unlock new markets.

Scalable Robotics Solutions for Real-world Applications

In dynamic environments where tasks and requirements can change rapidly, traditional robotics programming often falls short, meaning automation can’t effectively address most of the market's use cases. Additionally, operating robotic arms typically demands a high level of expertise.

Mbodi's system enables real-time adaptation by integrating and orchestrating intelligent generative AI agents with proven technologies like computer vision and path planning, alongside innovative communication patterns. This approach empowers users to teach robots new skills effortlessly and adapt to changing conditions. With the ability to translate natural language into robot actions in just 0.5 seconds, the system supports on-the-fly task modifications and rapid adjustments to new instructions, effectively bridging the gap between laboratory innovation and practical factory applications.

Balancing Innovation with Practicality

We view generative AI as a powerful tool within a broader ecosystem of technologies. Our approach leverages AI's strengths in areas like dynamic adaptation while relying on established technologies where they excel.‍

• Generative AI unlocks new levels of adaptability, such as modifying a robot’s behavior on-the-fly through natural language alone.

• Classical Robotics as Tool Users provides powerful capabilities, such as computer vision and path planning.

• Dynamic Agent-Centric Communication enables real-time decision-making and increasingly complex behavior.

With these technologies and components combined, the system can easily scale across many different hardware platforms and use cases.

Industry Recognition: Our Partnership with ABB

We’re collaborating with ABB, a global leader in robotics manufacturing, following our victory in the 2024 AI Startup Challenge, where we were selected from over 100 companies across 34 countries. Together, we are deploying one of the first generative AI systems of its kind in real-world industrial settings, enabling robots to learn and adapt. This marks a major advancement in making robotic automation accessible to everyone, particularly for those seeking flexible solutions in high-mix, low-volume production environments.

ABB, one of the world’s leading suppliers of robotics and machine automation, has been at the forefront of helping industries achieve resilience, flexibility, and efficiency, with a comprehensive portfolio of robots and automation solutions. By combining ABB’s decades of expertise in robotics with Mbodi's AI platform, we are accelerating the development of adaptive, next-generation automation solutions worldwide.

Open-Source

Meanwhile, we remain committed to open-source collaboration and lowering barriers to advanced AI technology in robotics. We open-sourced one of our libraries, embodied-agents, to enable seamless integration of state-of-the-art transformer models into robotics stacks. Additionally, we plan to release more open-source projects in the near future.

A Vision for Inclusive Robotics

Our mission at Mbodi is to make AI robotics technology accessible and beneficial across a wide range of industries and applications. By focusing on practical, agent-based solutions, we aim to address real-world challenges and drive meaningful progress in the field of robotics.

As this trend evolves, we anticipate a shift where agents won’t just solve predefined problems but will also determine how to approach new challenges independently. This evolution represents the future of autonomous systems, and it’s the kind of technology we’re developing at Mbodi – systems capable of adapting, learning, and acting independently to address diverse challenges in robotics and automation.

‍

‍