Agentic data science with the Data Agent Kit
cloud.google.com·7h ago
AI Research
arXiv CS.AI·Aug 28·5 min read
MOMO: A framework for seamless physical, verbal, and graphical robot skill learning and adaptation
TL;DR
MOMO is an innovative framework that enables flexible robot skill adaptation through kinesthetic touch, natural language, and a graphical interface. It successfully integrates these modalities, demonstrating practical applications in industrial settings, such as voice-commanded surface finishing.
✦ Why It Matters
Explore MOMO to enhance your robot's adaptability for diverse industrial tasks today.
Key Takeaways
- Implement kinesthetic touch for precise robot adjustments.
- Utilize natural language commands for high-level task modifications.
- Leverage graphical interfaces for visualizing robot trajectories.
- Explore voice-command capabilities for surface finishing tasks.
How It Works
MOMO operates by integrating three modalities: kinesthetic touch allows users to physically guide the robot, natural language enables high-level task instructions, and a graphical interface provides a visual representation of robot movements. The tool-based LLM architecture ensures safe adaptations by selecting and parameterizing predefined functions instead of generating code, which minimizes errors.
Kernelized Movement Primitives (KMPs) encode the robot's motion patterns, while probabilistic Virtual Fixtures assist in recording guided demonstrations. This combination allows for seamless transitions between different interaction methods, enhancing user experience and adaptability.